agentic-stack v0.19.1

agentic-stack v0.19.1

Patch release. Four correctness fixes in memory retrieval, project upgrade, and
file I/O. No contract, CLI, or on-disk format changes.

Fixes

• Recall returned superseded lessons (#60). Supersession appends a new id
  and never edits the old row, so retrieval kept surfacing retired guidance
  beside its replacement. superseded_by_map() now lives in
  render_lessons.py and is shared with recall.py, so rendering and
  retrieval can't disagree about what is retired. Provisional supersessions
  still do not retire the old lesson.
• upgrade copied new loop skills to a doubled path (#63). A missing
  loop-* skill landed at .agent/skills/skills/loop-x/ instead of
  .agent/skills/loop-x/, in both --dry-run reports and real copies.
• Locale-dependent encoding in learn.py (#61). stdout/stderr forced to
  UTF-8 and candidate JSON written with an explicit encoding, so non-ASCII
  claims survive on hosts whose default codepage isn't UTF-8.
• Leaked file handle (#62). _lesson_already_appended() reads through a
  context manager instead of relying on refcounting.

Upgrade

agentic-stack upgrade --dry-run
agentic-stack upgrade --yes

Optional — nothing breaks on 0.19.0. Projects that previously upgraded into
.agent/skills/skills/ can delete that stray directory; the correctly-placed
skills are copied on the next upgrade.

Verification

• python3 -m pytest: 180 passed (175 on 0.19.0 plus 5 new regression
  tests covering supersession filtering and the loop-skill copy path).
• Each fix ships with a test that fails on the pre-fix code.

Loop usage, safety boundaries, and the sandbox caveat are unchanged from
v0.19.0.

agentic-stack v0.19.0

agentic-stack v0.19.0

v0.19.0 adds portable agentic loops and a bounded local meta-harness.

Highlights

• JSON contracts for L1 report loops and L2/L3 action loops.
• Maker → deterministic verifier → independent checker feedback lifecycle.
• Atomic resumable checkpoints, privacy-safe local events, policy gates,
  stagnation detection, finite budgets, and owned Git worktrees.
• loop init, validate, run, resume, status, stop, cleanup, and
  audit commands through both agentic-stack and ./install.sh.
• Read-only loop summaries in doctor, status, dashboard, Mission Control, and
  the local data-layer export.

Usage

agentic-stack loop init /path/to/project
agentic-stack loop validate /path/to/project
agentic-stack loop run ci-sweeper "make the failing test green" /path/to/project --yes
agentic-stack loop status /path/to/project

Use agentic-stack upgrade --dry-run followed by agentic-stack upgrade --yes
to add loop starters to an existing installed project without overwriting
authored contracts or runtime state.

Safety boundaries

L2/L3 loops run mutating makers in owned worktrees, enforce deny-path and file
count gates, require deterministic verification and independent checker
approval, and stop on pause, budget, timeout, or repeated-failure decisions.
Events are local-only and whitelist metadata; prompts, tasks, command arguments,
stdout, stderr, and raw run IDs are not exported. The supervisor is not an operating-system sandbox; use harness-native sandboxes for stronger isolation.
Schedulers should execute one bounded command and inspect its exit status before
starting another.

Verification and attribution

Fresh verification evidence before tagging:

• python3 -m pytest: 166 passed.
• python3 -m harness_manager.cli loop validate --target . --json: valid
  ci-sweeper L2 contract.
• python3 -m harness_manager.cli loop audit --target . --json: valid audit,
  with zero real run evidence in the source checkout as expected.
• ./install.sh loop validate --target .: valid text dispatch.
• Fake-harness smoke: two-attempt completion, cleanup exit code 0, and the
  active checkout remained unchanged.
• The before/after git status --porcelain snapshots around the full suite
  were identical; the Claude hook suite ran against a session temp copy.

The loop design was informed by public loop-engineering discussions about
bounded feedback, verifier gates, resumability, and stagnation breakers. This
project implements its own Python standard-library contracts and does not copy
code.

agentic-stack v0.18.0

Highlights

• Adds agentic-stack brain ... as a first-class, optional bridge to the external codejunkie99/brain CLI/MCP memory system.
• Installs .agent/tools/brain_bridge.py into projects so host agents can explicitly query or write cross-harness long-term memory.
• Adds a brain seed skill with guidance for recall, durable notes, MCP setup, and secret avoidance.
• Keeps Brain external: no Rust workspace vendoring and no hard Homebrew dependency.

Commands

agentic-stack brain status
agentic-stack brain onboard --agents codex,cursor --yes
agentic-stack brain ask "auth decisions"
agentic-stack brain note "Use PKCE for local OAuth flows."
agentic-stack brain doctor --deep
agentic-stack brain tui
agentic-stack brain mcp-command

Upgrade Notes

Existing agentic-stack projects can pick up the new project bridge and skill with:

agentic-stack upgrade --dry-run
agentic-stack upgrade --yes

Brain itself remains a separate install. If it is not already available:

brew install codejunkie99/tap/brain

Validation

• PYTHONDONTWRITEBYTECODE=1 python3 -m pytest -q -> 95 passed
• PYTHONDONTWRITEBYTECODE=1 python3 -m pytest -q tests/test_brain_integration.py tests/test_upgrade_manifest_doctor.py -> 9 passed
• PYTHONDONTWRITEBYTECODE=1 python3 -m py_compile harness_manager/brain.py harness_manager/cli.py .agent/tools/brain_bridge.py
• PYTHONDONTWRITEBYTECODE=1 python3 -m harness_manager.schema
• bash -n install.sh
• ruby -c Formula/agentic-stack.rb
• git diff --check
• Temp-project smoke through ./install.sh claude-code, ./install.sh brain status, ./install.sh brain ask, and installed brain_bridge.py note using a fake Brain binary

agentic-stack v0.17.0

Minor release clearing the open PR queue and shipping the new adapter, Mission Control, and memory-review surfaces.

Highlights

• GitHub Copilot CLI adapter. Adds AGENTS.md, .github/instructions/, .github/hooks/, and .github/skills/ wiring so Copilot CLI can load the portable brain and run memory hooks.
• Google Gemini CLI adapter. Adds gemini.md plus a .gemini/skills/ mirror for Gemini CLI projects.
• Mission Control beta. Adds agentic-stack mission-control with a local web dashboard and static snapshot renderer.
• Semantic lesson retraction. Adds .agent/tools/retract_lesson.py so obsolete accepted lessons can be retracted from future recall while preserving append-only audit history.
• Test layout cleanup. Moves tests under tests/ with pytest coverage for adapters, upgrades, Mission Control, onboarding, and lesson retraction.

Upgrade

brew update
brew upgrade agentic-stack

For already-installed projects:

agentic-stack upgrade --dry-run
agentic-stack upgrade --yes
agentic-stack sync-manifest

Validation

• PYTHONDONTWRITEBYTECODE=1 python3 -m pytest -q -> 89 passed
• PYTHONDONTWRITEBYTECODE=1 python3 -m harness_manager.schema
• bash -n install.sh
• ruby -c Formula/agentic-stack.rb
• v0.17.0 tarball sha256: 704f8e7f05123b3791187e16f352936199e5e57e6855564c773961900ea13dd6

v0.16.1 — getting-started refresh

Patch release for PR #49 and the onboarding banner version fix.

Changed

• Refreshes docs/getting-started.md to separate Homebrew, source checkout, and PowerShell install flows.
• Documents the current management verbs for Homebrew and source-checkout users.

Fixed

• Onboarding banner now derives its displayed version from harness_manager.version instead of stale text.

v0.16.0 — safe project upgrades

Summary

• Add agentic-stack upgrade with dry-run and confirmed apply modes for safe installed-project infrastructure updates.
• Add agentic-stack sync-manifest and automatic manifest resync during install/add when .agent/skills exists.
• Make doctor warn for missing Claude Code hook command files and unwired hook scripts.

Homebrew

• Formula updated on master to v0.16.0.
• Tarball sha256: 55ffef80e990f1ceed6ec8016d66e8bab8b328762b3f5a3fdd80375dfa715dae

Verification

• python3 -m pytest -q -> 68 passed
• python3 -m py_compile harness_manager/cli.py harness_manager/install.py harness_manager/doctor.py harness_manager/skill_manifest.py harness_manager/upgrade.py
• python3 -m harness_manager.schema
• bash -n install.sh
• ruby -c Formula/agentic-stack.rb
• agentic-stack dashboard --plain smoke on the release worktree

v0.15.0 — dashboard TUI

[0.15.0] — 2026-05-06

Minor release. Adds a production dashboard TUI for installed agentic-stack
projects, with the trust-console inspection surface folded into the same
user-facing entrypoint.

Added

• agentic-stack dashboard / dash. Adds a terminal dashboard for project
  health, installed adapters, doctor checks, harness verification, memory, team
  brain, skills, managed instances, transfer, and local data exports.
• Trust-console parity. The dashboard includes a per-harness verify matrix,
  team brain status/init, skills listing, active instance listing,
  accepted/rejected memory review, and memory_why() evidence lookup.
• Plain renderer. agentic-stack dashboard --plain and non-TTY fallback
  produce a script-safe text dashboard for logs, agents, and tests.
• Interactive dashboard coverage. Adds local tests for renderer output,
  CLI aliases, trust-console parity sections, non-TTY fallback, up/down
  navigation, refresh, quit, and Enter-open behavior.

Changed

• Bare interactive agentic-stack / ./install.sh opens the dashboard when an
  existing .agent/install.json is present. Non-TTY shells keep printing
  command guidance instead of launching an interactive UI.
• README, POSIX installer help, and PowerShell installer help now document the
  dashboard entrypoint.

Fixed

• test_claude_code_hook.py now works both as a standalone validation script
  and as a pytest-collected module by providing a real mod fixture.

Migration

No migration required. Existing projects can run agentic-stack dashboard
after upgrading. Re-run an adapter install only if you want the latest copied
adapter guidance files in a project.

Release

• Tag v0.15.0 cut from master.
• GitHub release: https://github.com/codejunkie99/agentic-stack/releases/tag/v0.15.0
• Formula/agentic-stack.rb bumped to v0.15.0 in a follow-up commit after
  the tag tarball existed and its sha256 could be computed.
• Tarball sha256:
  e3fe0dde7a9997086a378123a365eced5514ad1a68871b294195fbf514611131.

v0.14.0

Minor release. Generalizes ztk from a Claude Code-only PreToolUse integration
into a host-neutral policy layer that works across hook-native, permission-rule,
and prompt-only coding tools.

Added

• Host-neutral ztk policy. Adds .agent/harness/ztk_policy.py and
  .agent/tools/ztk.py so one shared policy can evaluate shell commands,
  network fetches, file writes, and permission requests across harnesses.
• Claude Code and Codex hook adapters. Claude Code now routes PreToolUse
  events through ztk, while Codex installs opt-in hook configuration and handles
  PreToolUse, PermissionRequest, and PostToolUse events through the same policy.
• Portable ztk exec path. Cursor, Windsurf, OpenCode, OpenClaw, Hermes,
  Pi, and standalone agents now document python3 .agent/tools/ztk.py exec --
  as the common way to run policy-covered shell commands when the host has no
  equivalent native hook surface.
• Adapter capability matrix. Adds docs/adapter-capabilities.md with
  official-source notes for Claude Code hooks, Codex hooks, OpenCode
  permissions, Cursor rules, and Windsurf rules/memories.

Changed

• OpenCode adapter config now uses the current permission key.
• Codex installs .codex/config.toml and .codex/hooks.json through the
  manifest-driven adapter path instead of relying on installer-specific cases.
• README and per-harness docs now describe ztk as a portable policy layer, not
  a Claude Code-only hook.

Fixed

• Protected .agent/protocols/permissions.md edits are blocked by policy,
  including apply_patch payloads that mention the protected path.
• The shared pre-tool hook checks never-allowed patterns before approval-only
  patterns so hard denials cannot be downgraded to approval prompts.

Migration

No migration required. Re-run the relevant adapter install to copy the new hook,
permission, or prompt guidance files into an existing project.

v0.13.0

Minor release. Adds an onboarding-style transfer wizard for moving a portable
.agent brain into Codex, Cursor, Windsurf, or terminal-only projects with a
generated curl/PowerShell import command.

Added

• agentic-stack transfer wizard. Adds an onboarding-style TUI that parses
  natural-language requests such as move my memory into Codex, previews the
  target adapter files, asks for confirmation, and either generates an import
  command, applies locally, or both.
• Portable transfer bundles. Adds canonical JSON + gzip + base64url
  bundles with SHA-256 verification. The importer merges preferences and
  accepted lessons idempotently, restores selected memory files, copies skills,
  records import metadata, and installs selected adapters through the existing
  harness manager.
• Full memory intent. move my memory now means preferences, accepted
  lessons, skills, working memory, episodic/history logs, and candidate
  lessons. Data-layer exports, flywheel traces, runtime indexes, and caches
  stay out unless future scopes explicitly add them.
• Curl and PowerShell bootstraps. Adds scripts/import-transfer.sh and
  scripts/import-transfer.ps1 so another terminal can import a transfer
  bundle without manually cloning the repo first.

Changed

• Windsurf installs a modern .windsurf/rules/agentic-stack.md workspace rule
  and still writes legacy .windsurfrules for older Windsurf builds.
• agentic-stack transfer export and agentic-stack transfer import provide
  non-interactive surfaces for scripts and CI-style handoff flows.

Fixed

• Transfer export blocks secret-like content, including private keys and common
  API token patterns, before payload generation.
• Fresh Codex imports now copy the full .agent brain before installing the
  Codex AGENTS.md and .agents/skills adapter wiring.

Migration

No migration required. Existing installs keep working. Run
agentic-stack transfer from a project that already has .agent/ to create or
import a transfer bundle.

v0.12.0 — tldraw visual canvas

Minor release. Adds the opt-in tldraw seed skill for live canvas diagrams
and a skill-local snapshot store. The feature stays beta and off by default.

Added

• tldraw seed skill — live canvas diagrams. Adds
  .agent/skills/tldraw/SKILL.md with MCP tool guidance, shape constraints,
  and a self-rewrite hook for diagram, sketch, wireframe, flowchart,
  whiteboard, and architecture visualization prompts.
• Skill-local snapshot store. Adds .agent/skills/tldraw/store.py with
  snapshot, list, load, and archive CLI/API support. Runtime output is
  local and gitignored under .agent/skills/tldraw/: snapshots.jsonl,
  snapshots/, and INDEX.md.
• Opt-in feature flag. Onboarding now writes a tldraw beta feature flag,
  default off. The skill loader skips flagged skills unless
  .agent/memory/.features.json explicitly enables them.
• Manual MCP config source. Adds adapters/_shared/tldraw-mcp.json as the
  canonical tldraw MCP config block users can merge into Claude Code,
  Cursor, or Antigravity after enabling the feature.

Changed

• Seed skill count is now nine: skillforge, memory-manager, git-proxy,
  debug-investigator, deploy-checklist, design-md, data-layer,
  data-flywheel, and tldraw.
• tldraw persistence is intentionally skill-local storage, not a fifth memory
  layer: it has no dream-cycle, clustering, recall, or semantic-memory
  lifecycle.
• Default adapter installs no longer wire beta tldraw MCP config
  automatically.

Fixed

• Updated the tldraw validation suite to target
  .agent/skills/tldraw/store.py instead of the removed
  .agent/memory/visual/visual_memory.py.
• store.py renders INDEX.md while holding the JSONL lock so concurrent
  snapshots cannot leave the index stale.
• Added coverage for feature-gated skill loading, no default MCP install, path
  traversal rejection, malformed JSONL recovery, same-second snapshot ids, and
  concurrent snapshots.

Migration

No migration required. Existing installs keep tldraw disabled until users opt
in through onboarding reconfiguration or .agent/memory/.features.json.

Credits

• PR #11 by @Siddharth11Roy, with review fixes and release prep by Codex.

Release

• Tag v0.12.0 cut from master.
• GitHub release: https://github.com/codejunkie99/agentic-stack/releases/tag/v0.12.0
• Formula/agentic-stack.rb bumped to v0.12.0 in a follow-up commit after
  the tag tarball existed and its sha256 could be computed.
• Tarball sha256:
  cd18a0cf82e027390ef10a102ec4bfed26fc45114f6ce96c0787e5a1cf0dce26.