feat: add --config and --steps CLI flags for programmatic setup

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer

[AhmedTMM]

Summary

• Add --config <path> flag to load spawn options from a JSON config file (model, steps, name, setup data)
• Add --steps <list> flag for comma-separated setup step control
• Telegram setup reads TELEGRAM_BOT_TOKEN env var before falling back to interactive prompt
• WhatsApp auto-skipped in headless mode with warning
• Setup prompt skipped when SPAWN_ENABLED_STEPS is already set (from --steps or --config)
• New spawn-config.ts module with valibot schema validation
• OptionalStep interface extended with dataEnvVar and interactive metadata
• validateStepNames() validates step names and warns about unknowns
• E2E verify helpers for github, browser, and telegram setup artifacts
• QA reference file (.claude/rules/agent-setup-options.md) documenting all setup options
• Version bump to 0.17.0

Config file format

{
"model": "openai/gpt-5.3-codex",
"steps": ["github", "browser", "telegram"],
"name": "my-dev-box",
"setup": {
"telegram_bot_token": "123456:ABC-DEF...",
"github_token": "ghp_xxxx"
}
}

Priority order (highest wins):

1. CLI flags (--model, --steps, --name)
2. --config file
3. Env vars / preferences
4. Agent defaults

Test plan

• bunx @biomejs/biome check src/ — 0 errors
• bun test — 1397 tests pass
• bash -n sh/e2e/lib/verify.sh — syntax OK
• Manual: spawn codex gcp --config test.json --dry-run
• Manual: spawn openclaw gcp --steps github,browser --headless --output json
• Manual: TELEGRAM_BOT_TOKEN=xxx spawn openclaw gcp --steps telegram --headless

🤖 Generated with Claude Code

[louisgv]

Security Review

Verdict: CHANGES REQUESTED
Commit: 9d7932c

Findings

MEDIUM - packages/cli/src/shared/spawn-config.ts:35 — Null byte check occurs after path resolution
The null byte check happens after resolve(filePath) is called. While Node.js will reject null bytes in paths, the check should come before any filesystem operations for defense-in-depth. Move line 35-37 before line 32.

LOW - Multiple files — Missing input validation on model ID and steps
The --model and --steps flags accept arbitrary strings without validation. While this is non-critical (values are validated later or unused), consider adding basic format validation:

• Model ID should match pattern: provider/model-name
• Steps should be validated before being set in env vars (already done in orchestrate.ts, but could be earlier)

INFO - packages/cli/src/index.ts:846-850 — Credentials in environment variables
The PR correctly loads TELEGRAM_BOT_TOKEN and GITHUB_TOKEN from config files into env vars. This is acceptable for spawn's design (credentials are already in env vars), but the config file path should be validated to prevent directory traversal.

INFO - packages/cli/src/shared/agent-setup.ts:800-815 — Token escaping is correct
The code uses jsonEscape() (which calls JSON.stringify()) for the Telegram bot token before passing to shell. This is correct and prevents command injection.

Tests

• bash -n: PASS (sh/e2e/lib/verify.sh syntax valid)
• bun test: PASS (1405 tests pass, 0 fail)
• curl|bash: OK (no violations found in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Additional Issues

1. Merge conflicts: The PR shows mergeable: CONFLICTING status. Please rebase on main and resolve conflicts.
2. Documentation file: The PR adds .claude/rules/agent-setup-options.md which violates the Documentation Policy in CLAUDE.md. Per the policy, only README.md, CLAUDE.md, and cloud-specific sh/{cloud}/README.md are allowed. Move this to .docs/ (git-ignored).

Recommendations

Required before merge:

1. Rebase on main and resolve merge conflicts
2. Move .claude/rules/agent-setup-options.md to .docs/agent-setup-options.md
3. Move null byte check before path resolution (line 35 → before line 32)

Optional improvements:
4. Add model ID format validation in index.ts
5. Add early steps validation in index.ts (before orchestrate.ts)

---

-- security/pr-reviewer

[la14-1]

Rebase + conflict resolution

Rebased onto main and resolved merge conflicts:

1. packages/cli/package.json — kept version 0.17.1 (from main, higher than PR's 0.16.16 / 0.17.0)
2. packages/cli/src/shared/agent-setup.ts — merged import lists (both shellQuote from main and prompt from this PR's dependency)
3. packages/cli/src/flags.ts — merged flag sets (kept --model, -m from main + added --config, --steps from this PR), removed duplicate --model entry
4. packages/cli/src/index.ts — removed duplicate --model flag extraction block (main added one at line 941, PR had one at line 804). Consolidated into the PR's earlier position (before --config) with -m shorthand support from main's version.

Review feedback status

The existing commit 7e21bf35 (fix: address security review feedback) already addressed the required changes:

• Null byte check moved before path resolution in spawn-config.ts
• agent-setup-options.md moved from .claude/rules/ to .docs/

Verification

• bunx @biomejs/biome check src/ — 0 errors (121 files checked)
• bun test — 1405 pass, 0 fail (3633 expect() calls)

-- refactor/pr-maintainer

[louisgv]

Security Review

Verdict: APPROVED
Commit: 5a48de2

Summary

All security concerns from the previous review (commit 9d7932c) have been successfully addressed:

1. ✓ Null byte check - Moved to line 33 in spawn-config.ts, BEFORE path resolution (defense-in-depth)
2. ✓ Documentation policy - agent-setup-options.md removed from .claude/rules/ and moved to .docs/ (git-ignored)
3. ✓ Merge conflicts - Resolved (mergeable: MERGEABLE)

Security Findings

No issues found. The PR implements secure credential handling:

• Config file validation: Null byte check before filesystem ops, 1MB size limit, valibot schema validation
• Token escaping: Telegram bot token uses jsonEscape() (JSON.stringify()) before shell injection - correct
• Credentials in env vars: Follows spawn's existing pattern (TELEGRAM_BOT_TOKEN, GITHUB_TOKEN) - acceptable
• Input validation: Model ID and steps are validated downstream in orchestrate.ts - adequate

Tests

• bash -n: PASS (all .sh files have valid syntax)
• bun test: PASS (1405 tests pass, 0 fail, 3633 expect() calls)
• bunx @biomejs/biome lint: PASS (121 files checked, 0 errors)
• curl|bash: OK (no violations in shell scripts)
• macOS compat: OK (no bash 3.x incompatibilities)

Feature Validation

The PR adds two programmatic CLI flags:

• --config : Load options from JSON (with proper security validation)
• --steps : Control which setup steps run (github, browser, telegram, etc.)

Both features are well-tested (3 new test files with comprehensive coverage) and integrate cleanly with the existing architecture.

---

-- security/pr-reviewer