feat: enhanced testing

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"

.claude/skills/setup-agent-team/discovery.sh

@@ -201,12 +201,25 @@ For each candidate, verify:
 - Has affordable CPU instances (pay-per-hour or pay-per-second pricing)
 - Is actually available (not waitlisted/invite-only)
 
-**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update:
-1. `test/record.sh` — add the cloud to `ALL_RECORDABLE_CLOUDS`, add a case in `get_endpoints()` with the cloud's GET endpoints, add a case in `get_auth_env_var()`, add a case in `call_api()`, and add a `_live_{cloud}()` function for create/delete fixture recording
-2. `test/mock.sh` — add a URL-stripping case in the curl mock so the cloud's API base URL is recognized (look for the `case "$URL" in` block near line 133)
-3. `test/record.sh` `has_api_error()` — add error detection for the cloud's API error format
-
-Without these, the new cloud will have no test coverage and the QA cycle will skip it entirely.
+**MANDATORY: Add new clouds to the test infrastructure.** When adding a new cloud, you MUST also update ALL of these files:
+
+**`test/record.sh`** (fixture recording):
+1. Add the cloud to `ALL_RECORDABLE_CLOUDS`
+2. Add a case in `get_endpoints()` with the cloud's GET-only API endpoints
+3. Add a case in `get_auth_env_var()` mapping the cloud to its auth env var
+4. Add a case in `call_api()` to dispatch to the cloud's API function
+5. Add a case in `has_api_error()` for the cloud's error response format
+6. Add a `_live_{cloud}()` function for create/delete server fixture recording
+
+**`test/mock.sh`** (mock test infrastructure — ALL of these):
+1. **URL stripping**: Add a case in `_strip_api_base()` to recognize the cloud's API base URL (e.g., `https://api.newcloud.com/v1*`). Without this, mock curl logs `UNHANDLED_URL` warnings.
+2. **Body validation**: Add a case in `_validate_body()` with the cloud's server creation endpoint and required POST body fields (e.g., `newcloud) case "$EP_CLEAN" in /instances) _check_fields "name region type" ;; esac ;;`)
+3. **Cloud API assertions**: Add a case in `assert_cloud_api_calls()` with the cloud's expected API calls (SSH key fetch endpoint + server create endpoint)
+4. **Environment setup**: Add a case in `setup_env_for_cloud()` with the cloud's test env vars (API token, server name, plan, region, etc.)
+5. **POST response**: If the cloud's create-server response format differs from the default, add a case in `_respond_post()` and `_synthetic_active_response()`
+6. **State tracking**: If the cloud creates servers at a non-standard endpoint, add it to `_track_state()`
+
+Without ALL of these, the new cloud will have no test coverage, body validation will be missing, and the QA cycle will skip it entirely.
 
 ### Agent Scout (spawn 1, only if justified)
 Research new AI agents, BUT only add one if there's REAL community demand:
@@ -474,9 +487,13 @@ providers — container platforms, budget VPS providers, or regional clouds with
 REST APIs. We need affordable instances for running agents that use remote API inference, NOT GPU clouds.
 Create lib/common.sh, add to manifest, implement 2-3 agents, add "missing" entries for the rest.
 
-MANDATORY: When adding a new cloud, also add it to the test infrastructure:
+MANDATORY: When adding a new cloud, also update the test infrastructure:
 1. test/record.sh — add to ALL_RECORDABLE_CLOUDS, get_endpoints(), get_auth_env_var(), call_api(), has_api_error(), and add a _live_{cloud}() function
-2. test/mock.sh — add a URL-stripping case in the curl mock (case "$URL" in block)
+2. test/mock.sh — add ALL of these:
+ - _strip_api_base(): URL-stripping case for the cloud's API base URL
+ - _validate_body(): body validation case with server creation endpoint + required fields
+ - assert_cloud_api_calls(): expected API calls (SSH key fetch + server create)
+ - setup_env_for_cloud(): test env vars (API token, server name, plan, region)
 
 Only add a new AGENT if you find one with real community buzz:
 - 1000+ GitHub stars

.claude/skills/setup-agent-team/qa-cycle.sh

@@ -172,21 +172,15 @@ for branch in $REMOTE_README_BRANCHES; do
  fi
 done
 
-# Close stale qa PRs (open > 2 hours)
+# Close stale qa PRs (open > 2 hours) — never auto-merge, leave for human review
 STALE_PRS=$(gh pr list --state open --label '' --json number,headRefName,updatedAt \
  --jq '[.[] | select(.headRefName | startswith("qa/")) | select(.updatedAt < (now - 7200 | todate)) | .number] | .[]' 2>/dev/null) || true
 for pr_num in $STALE_PRS; do
  if [[ -n "$pr_num" ]]; then
- PR_MERGEABLE=$(gh pr view "$pr_num" --json mergeable --jq '.mergeable' 2>/dev/null) || PR_MERGEABLE="UNKNOWN"
- if [[ "$PR_MERGEABLE" == "MERGEABLE" ]]; then
- log "Merging stale QA PR #${pr_num}..."
- gh pr merge "$pr_num" --squash --delete-branch 2>&1 | tee -a "${LOG_FILE}" || true
- else
- log "Closing unmergeable stale QA PR #${pr_num}..."
- gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
+ log "Closing stale QA PR #${pr_num}..."
+ gh pr close "$pr_num" --comment "Auto-closing: stale QA PR from a previous cycle.
 
 -- qa/cycle" 2>&1 | tee -a "${LOG_FILE}" || true
- fi
  fi
 done
 
@@ -228,8 +222,14 @@ check_timeout || exit 0
 log "=== Phase 1: Record fixtures ==="
 
 RECORD_OUTPUT="/tmp/spawn-qa-record-output.txt"
+RECORD_FAILURES_FILE="${REPO_ROOT}/.docs/qa-record-failures.json"
 rm -f "${RECORD_OUTPUT}"
 
+# Initialize failure tracker if missing
+if [[ ! -f "${RECORD_FAILURES_FILE}" ]]; then
+ printf '{}' > "${RECORD_FAILURES_FILE}"
+fi
+
 RECORD_EXIT=0
 bash test/record.sh allsaved 2>&1 | tee -a "${LOG_FILE}" | tee "${RECORD_OUTPUT}" || RECORD_EXIT=$?
 
@@ -408,6 +408,107 @@ Only modify ${cloud}/lib/common.sh and test/record.sh if the recording infrastru
  fi
 fi
 
+# --- Track consecutive Phase 1 failures per cloud ---
+# Parse the final recording output to determine which clouds failed
+FINAL_RECORD_FAILED=""
+FINAL_RECORD_SUCCEEDED=""
+if [[ -f "${RECORD_OUTPUT}" ]]; then
+ _current_cloud=""
+ _cloud_had_error=""
+ while IFS= read -r line; do
+ clean=$(printf '%s' "$line" | sed 's/\x1b\[[0-9;]*m//g')
+ case "$clean" in
+ *"Recording "*" ━━━"*)
+ # Save previous cloud result
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+ _current_cloud=$(printf '%s' "$clean" | sed 's/.*Recording //; s/ ━━━.*//')
+ _cloud_had_error=""
+ ;;
+ *"fail "*)
+ _cloud_had_error="true"
+ ;;
+ *"done "*)
+ # Cloud section ended
+ ;;
+ esac
+ done < "${RECORD_OUTPUT}"
+ # Handle last cloud
+ if [[ -n "${_current_cloud}" ]]; then
+ if [[ "${_cloud_had_error}" == "true" ]]; then
+ FINAL_RECORD_FAILED="${FINAL_RECORD_FAILED} ${_current_cloud}"
+ else
+ FINAL_RECORD_SUCCEEDED="${FINAL_RECORD_SUCCEEDED} ${_current_cloud}"
+ fi
+ fi
+fi
+FINAL_RECORD_FAILED=$(printf '%s' "${FINAL_RECORD_FAILED}" | sed 's/^ //')
+FINAL_RECORD_SUCCEEDED=$(printf '%s' "${FINAL_RECORD_SUCCEEDED}" | sed 's/^ //')
+
+# Update the persistent failure tracker
+if [[ -f "${RECORD_FAILURES_FILE}" ]]; then
+ python3 -c "
+import json, sys
+
+tracker_path = sys.argv[1]
+failed = sys.argv[2].split() if sys.argv[2] else []
+succeeded = sys.argv[3].split() if sys.argv[3] else []
+
+try:
+ with open(tracker_path) as f:
+ tracker = json.load(f)
+except (json.JSONDecodeError, FileNotFoundError):
+ tracker = {}
+
+# Increment consecutive failures for failed clouds
+for cloud in failed:
+ tracker[cloud] = tracker.get(cloud, 0) + 1
+
+# Reset counter for clouds that succeeded
+for cloud in succeeded:
+ tracker[cloud] = 0
+
+with open(tracker_path, 'w') as f:
+ json.dump(tracker, f, indent=2, sort_keys=True)
+
+# Output clouds that hit the threshold (3+ consecutive failures)
+escalate = [c for c, count in tracker.items() if count >= 3]
+if escalate:
+ print(' '.join(escalate))
+" "${RECORD_FAILURES_FILE}" "${FINAL_RECORD_FAILED}" "${FINAL_RECORD_SUCCEEDED}" > /tmp/spawn-qa-escalate.txt 2>/dev/null || true
+
+ ESCALATE_CLOUDS=$(cat /tmp/spawn-qa-escalate.txt 2>/dev/null || true)
+ rm -f /tmp/spawn-qa-escalate.txt
+
+ if [[ -n "${ESCALATE_CLOUDS}" ]]; then
+ for cloud in ${ESCALATE_CLOUDS}; do
+ consecutive=$(python3 -c "import json; print(json.load(open('${RECORD_FAILURES_FILE}')).get('${cloud}', 0))" 2>/dev/null || printf "3+")
+ log "Phase 1: ESCALATION — ${cloud} has failed ${consecutive} consecutive cycles"
+
+ # Check if an issue already exists for this cloud
+ existing_issue=$(gh issue list --repo OpenRouterTeam/spawn --state open \
+ --search "fixture recording failing ${cloud}" \
+ --json number --jq '.[0].number' 2>/dev/null) || existing_issue=""
+
+ if [[ -z "${existing_issue}" ]]; then
+ gh issue create --repo OpenRouterTeam/spawn \
+ --title "QA: ${cloud} fixture recording has been failing for ${consecutive} consecutive cycles" \
+ --body "$(printf 'The automated QA cycle has detected that fixture recording for **%s** has failed for **%s consecutive cycles**.\n\nThis likely indicates a persistent issue with the cloud provider'\''s API or our integration that requires manual investigation.\n\n## What to check\n- Has the %s API changed? (new auth requirements, endpoint changes, rate limits)\n- Are the API credentials still valid?\n- Check `%s/lib/common.sh` for outdated API calls\n- Run `bash test/record.sh %s` locally to reproduce\n\n## Auto-generated\nThis issue was created automatically by the QA cycle (`qa-cycle.sh`).\n\n-- qa/cycle' "${cloud}" "${consecutive}" "${cloud}" "${cloud}" "${cloud}")" \
+ --label "bug" \
+ 2>&1 | tee -a "${LOG_FILE}" || true
+ log "Phase 1: Created GitHub issue for ${cloud} persistent failure"
+ else
+ log "Phase 1: Issue #${existing_issue} already open for ${cloud}, skipping duplicate"
+ fi
+ done
+ fi
+fi
+
 rm -f "${RECORD_OUTPUT}"
 check_timeout || exit 0
 
@@ -448,6 +549,15 @@ else
  FAILED_CLOUDS=$(grep ':fail$' "${RESULTS_PHASE2}" | sed 's/:fail$//' | cut -d/ -f1 | sort -u || true)
  fi
 
+ # Capture full mock test output per-cloud for richer agent context
+ MOCK_OUTPUT_DIR="/tmp/spawn-qa-mock-output"
+ rm -rf "${MOCK_OUTPUT_DIR}"
+ mkdir -p "${MOCK_OUTPUT_DIR}"
+ for cloud in $FAILED_CLOUDS; do
+ log "Phase 3: Capturing full mock test output for ${cloud}..."
+ bash test/mock.sh "$cloud" > "${MOCK_OUTPUT_DIR}/${cloud}.log" 2>&1 || true
+ done
+
  AGENT_PIDS=""
  for cloud in $FAILED_CLOUDS; do
  check_timeout || break
@@ -456,25 +566,21 @@ else
  cloud_failures=$(printf '%s\n' $FAILURES | grep "^${cloud}/" || true)
  failing_scripts=""
  failing_agents=""
- error_context=""
  for combo in $cloud_failures; do
  agent=$(printf '%s' "$combo" | cut -d/ -f2)
  script_path="${cloud}/${agent}.sh"
  failing_scripts="${failing_scripts} ${script_path}"
  failing_agents="${failing_agents} ${agent}"
- if [[ -f "${LOG_FILE}" ]]; then
- ctx=$(grep -A 10 "test ${script_path}" "${LOG_FILE}" | tail -10 || true)
- if [[ -n "$ctx" ]]; then
- error_context="${error_context}
---- ${script_path} ---
-${ctx}
-"
- fi
- fi
  done
  failing_scripts=$(printf '%s' "$failing_scripts" | sed 's/^ //')
  failing_agents=$(printf '%s' "$failing_agents" | sed 's/^ //')
 
+ # Use full mock test output as error context (not just 10 lines)
+ error_context=""
+ if [[ -f "${MOCK_OUTPUT_DIR}/${cloud}.log" ]]; then
+ error_context=$(cat "${MOCK_OUTPUT_DIR}/${cloud}.log")
+ fi
+
  fail_count=$(printf '%s\n' $cloud_failures | wc -l | tr -d ' ')
  log "Phase 3: Spawning agent to fix ${fail_count} failing script(s) in ${cloud}"
 
@@ -582,6 +688,9 @@ FIXEOF
  done
  git worktree prune 2>/dev/null || true
 
+ # Clean up per-cloud mock output
+ rm -rf "${MOCK_OUTPUT_DIR}" 2>/dev/null || true
+
  log "Phase 3: Fix agents complete"
 fi
 

.claude/skills/setup-trigger-service/qa-cycle.sh

@@ -0,0 +1 @@
+../setup-agent-team/qa-cycle.sh

.github/workflows/test.yml

@@ -0,0 +1,28 @@
+name: Test
+
+on:
+ pull_request:
+ branches: [main]
+
+concurrency:
+ group: test-${{ github.head_ref }}
+ cancel-in-progress: true
+
+jobs:
+ mock-tests:
+ name: Mock Tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ steps:
+ - name: Checkout code
+ uses: actions/checkout@v4
+ - name: Run mock tests
+ run: |
+ bash test/mock.sh 2>&1 | tee /tmp/mock-output.log
+ - name: Post summary
+ if: always()
+ run: |
+ echo '## Mock Test Results' >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"
+ grep -E '(Results:|✓|✗|skip|━━━)' /tmp/mock-output.log | head -100 >> "$GITHUB_STEP_SUMMARY"
+ echo '```' >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,6 +2,7 @@ node_modules/
 .docs/
 __pycache__/
 .claude/skills/*/start-*.sh
+.claude/settings.local.json
 cli/cli.js
 
 # Sensitive files — never commit secrets or private keys
@@ -15,3 +16,5 @@ id_rsa
 id_ed25519
 credentials.json
 service-account.json
+.claude/settings.json
+.DS_Store

CLAUDE.md

@@ -244,6 +244,29 @@ macOS ships bash 3.2. All scripts MUST work on it:
 - Run tests with `bun test`
 - Use `import { describe, it, expect, beforeEach, afterEach, mock, spyOn } from "bun:test"`
 
+### Mock Test Infrastructure (MANDATORY for new clouds)
+
+When adding a new cloud provider, you **MUST** update all of these in `test/mock.sh`:
+
+| Function | What to add |
+|---|---|
+| `_strip_api_base()` | URL pattern to strip the cloud's API base (e.g., `https://api.newcloud.com/v1*`) |
+| `_validate_body()` | Server creation endpoint + required POST fields |
+| `assert_cloud_api_calls()` | Expected API calls (SSH key fetch + server create endpoints) |
+| `setup_env_for_cloud()` | Test env vars (API token, server name, plan, region) |
+
+And in `test/record.sh`:
+
+| Function | What to add |
+|---|---|
+| `ALL_RECORDABLE_CLOUDS` | Cloud name to the list |
+| `get_endpoints()` | GET-only API endpoints for fixture recording |
+| `get_auth_env_var()` | Auth env var name mapping |
+| `call_api()` | API function dispatcher |
+| `has_api_error()` | Error response detection for the cloud's API |
+
+Without these, the cloud has **no test coverage**, body validation is missing, mock tests log `UNHANDLED_URL` warnings, and the QA cycle skips the cloud entirely. Both `test/mock.sh` and `test/record.sh` run in parallel when testing/recording multiple clouds.
+
 ## CLI Version Management
 
 **CRITICAL: Bump the version on every CLI change!**

genesiscloud/claude.sh

@@ -20,9 +20,9 @@ create_server "${SERVER_NAME}"
 verify_server_connectivity "${GENESIS_SERVER_IP}"
 wait_for_cloud_init "${GENESIS_SERVER_IP}" 60
 
-log_step "Verifying Claude Code installation..."
+log_warn "Verifying Claude Code installation..."
 if ! run_server "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && command -v claude" >/dev/null 2>&1; then
-log_step "Claude Code not found, installing manually..."
+log_warn "Claude Code not found, installing manually..."
  run_server "${GENESIS_SERVER_IP}" "curl -fsSL https://claude.ai/install.sh | bash"
 fi
 log_info "Claude Code is installed"
@@ -34,7 +34,7 @@ else
  OPENROUTER_API_KEY=$(get_openrouter_api_key_oauth 5180)
 fi
 
-log_step "Setting up environment variables..."
+log_warn "Setting up environment variables..."
 inject_env_vars_ssh "${GENESIS_SERVER_IP}" upload_file run_server \
  "OPENROUTER_API_KEY=${OPENROUTER_API_KEY}" \
  "ANTHROPIC_BASE_URL=https://openrouter.ai/api" \
@@ -52,7 +52,7 @@ log_info "Genesis Cloud server setup completed successfully!"
 log_info "Server: ${SERVER_NAME} (ID: ${GENESIS_SERVER_ID}, IP: ${GENESIS_SERVER_IP})"
 echo ""
 
-log_step "Starting Claude Code..."
+log_warn "Starting Claude Code..."
 sleep 1
 clear
 interactive_session "${GENESIS_SERVER_IP}" "export PATH=\$HOME/.local/bin:\$PATH && source ~/.zshrc && claude"