Added

• Optional [semantic] extra (#11) — pip install aim-cli[semantic] adds embeddings (sentence-transformers) powering:
  • aim search --semantic — rank tasks/docs/memories by meaning, with graceful fallback to keyword search when the extra is absent.
  • an aim doctor similar-memory check — flags highly-similar memory pairs as possible duplicates/contradictions.
    The core stays strictly zero-dependency.

Docs

• Recorded the decision to freeze time-tracking / user management as maintenance-only (#12, ADR 0001).

Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.9.0
(with embeddings: pip install "aim-cli[semantic] @ git+https://github.com/phuonghx/aim-cli.git@v1.9.0")

Added

• Dashboard "Health" tab (#9) — surfaces aim doctor context-drift findings in the Control Hub via a token-protected /api/doctor endpoint, grouped by severity with suggested fixes.

Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.8.0

Two-way GitHub sync (#7) — completes the GitHub integration.

Added

• aim github pull [id] [--all] [--dry-run] — reconcile AIM tasks from their linked GitHub issues. GitHub is canonical for issue state (CLOSED→done, reopened→todo) and title; AIM stays canonical for acceptance criteria, dependencies, and spec links. --dry-run previews drift.
• aim github status --check — fetch live issues and report drift vs AIM.

Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.7.0

Backlog items from the AIM Roadmap project.

Added

• aim task renumber <old> <new> (#10) — rename a task and rewrite every reference (@task-N in tasks/docs, plus dependsOn/parent), word-boundary safe. Resolves the duplicate/mismatched-id findings from aim doctor.

Changed

• aim github push --project N (#8) — now sets each card's Project (v2) Status field to match its AIM status (todo→Todo, in-progress/in-review→In Progress, done→Done).

Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.6.0

Roadmap Phase 5 (final) — GitHub sync. Projects AIM tasks onto GitHub Issues and Projects so a team gets a familiar board; AIM stays the agent's working layer. Zero-dependency — shells out to the gh CLI.

Added

• aim github push [id] [--all] [--project N] — create/update a GitHub issue per task, idempotent via a stored **GitHub Issue:** number. Maps done → closed, otherwise open; optionally adds issues to a Project (v2).
• aim github status — show task ↔ issue linkage.
• aim github create-project <title> — create a Project (v2) for the repo owner.
• Tasks carry a githubIssue field.

Completes the AIM roadmap (Phases 0–5). Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.5.0

Roadmap Phase 4 — spec-driven development. Makes the existing spec/plan task fields meaningful and integrates with GitHub spec-kit.

Added

• aim spec import <dir> — imports a spec-kit feature directory (spec.md, optional plan.md) into .ai-context/docs/specs|plans and creates an umbrella task linked to the spec (expand via the decompose_prd MCP prompt or --depends-on).
• aim spec coverage — reports how many tasks have a linked spec.
• aim validate --require-spec — fails when any task has no linked spec (spec-driven CI gate).

Zero-dependency, backward-compatible. Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.4.0

Roadmap Phase 3 — task intelligence. The deterministic parts of an AI task manager, with no LLM dependency.

Added

• Task dependencies — tasks carry dependsOn (a **Depends On:** line). aim task create --depends-on N; aim task edit --add-dep / --remove-dep N, with cycle detection.
• aim task next / MCP next_task — the highest-priority, lowest-id task that is not done/blocked and whose dependencies are all done.
• PRD decomposition — MCP decompose_prd prompt instructs the connected agent to break a PRD into tasks and create them via the new batch tool create_tasks (resolves within-batch dependency chains via per-task key). The server now advertises the prompts capability.

Zero-dependency, backward-compatible. Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.3.0

Roadmap Phase 2 — reverse-sync. Adopt AIM without hand-copying your existing scattered rules.

Added

• aim ingest — collects hand-written rules from known files (every sync target plus .clinerules / .rules / .aider.conf.yml) into .ai-context/imported/*.md. Ignores AIM's own managed block, so it never re-imports generated output (idempotent). --dry-run previews; --emit hands raw content to the connected agent to restructure into config.json.

Changed

• aim sync now re-emits the consolidated imported rules into every client file under an "Imported Project Rules" section.

Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.2.0

Roadmap Phase 0 + Phase 1 — the "living context layer": a context layer that keeps itself fresh and closes the correction feedback loop.

Added

• aim doctor — deterministic, no-LLM context-drift detection: stale memories (cross-referenced against git history of the files they mention), broken refs, duplicate/mismatched task IDs, spec drift, idle tasks, spec coverage. Exits non-zero on high/medium findings (CI gate). --mine filters to your own memories.
• Correction loop — MCP record_correction(...) captures a mid-session correction as a memory that syncs to every tool. New MCP tools review_memory and doctor.
• Memory lifecycle — aim memory edit / rm / review (was add/list only); review resets the staleness clock. Records now carry author, reviewedAt, status, and auto-extracted refs.
• Real global memory layer — -l global persists to ~/.aim/memories.json and is merged into every project.

Install: pip install git+https://github.com/phuonghx/aim-cli.git@v1.1.0

First stable public release of AIM (AI Memory/Mind) — a centralized, stdlib-only context/task/memory manager for AI coding assistants. Earlier 0.x builds were internal pre-releases.

Install

pip install git+https://github.com/phuonghx/aim-cli.git@v1.0.0

or download the wheel below. Then aim init in your project; verify with aim --version.

Highlights

• Workspace CLI — tasks with subtasks/labels/priorities/acceptance-criteria (lossless markdown round-trip), docs library, persistent memory, search, reference validation, time tracking, ASCII Kanban board, users, and code-generation templates. aim demo seeds a sample workspace.
• AI-client sync — aim sync compiles one config into CLAUDE.md, AGENTS.md, GEMINI.md, ANTIGRAVITY.md, .cursor/rules/aim.mdc (+ legacy .cursorrules/.windsurfrules), and Copilot instructions. Non-destructive: content outside <!-- AIM:BEGIN/END --> markers is preserved.
• MCP server (aim mcp) — serves the workspace over the Model Context Protocol (stdio, zero dependencies) so assistants query/mutate tasks, docs, and memories directly. Register: claude mcp add aim -- aim mcp.
• Control Hub dashboard (aim browser) — Kanban with drag-and-drop, docs viewer, memory grid, time tracking, and dependency graph. Localhost-only with per-launch token + Host validation, path-traversal guards, and sanitized rendering.
• Reliability — atomic JSON writes with corrupt-store backups, race-safe task IDs, parent-cycle prevention, UTF-8 piped output, and aim init --force with backups.
• Quality — shared aim/core.py service layer, MIT license, 36-test pytest suite, and CI (ruff + pytest on Windows & Ubuntu, Python 3.9/3.13) gating releases.

See the full CHANGELOG.