project-management-plugin

Set or clear the focus anchor that keeps Claude on task across every turn

Autonomous project execution loop that runs until completion or block

Backlog grooming: reorder, refine, split, or archive tasks

Cap the number of tool calls allowed for the current task; nudges Claude at 80% and 120%

Save or restore project state checkpoints

Inspect or repair project state files and caches

Delegate a task to a specific specialized agent

Declare the binary-testable criteria Claude must verify before the session may end

Audit the breadcrumb trail for scope drift, over-engineering, and criterion advancement

Narrow execution scope to a specific phase, epic, or story

Snapshot active task state for post-/compact resume, or display the current handoff

8-phase interview-first project initialization with micro-task decomposition

Connect an external PM platform (Jira, Linear, Notion, Asana, Trello, ClickUp, Monday, Todoist, GitHub)

Show next actionable tasks based on dependencies and priority

Recursive task decomposition from phases to stories to micro-tasks (5-15 min)

Generate a full project report with progress, risks, and artifacts

Run deep research protocol for a specific task using codebase, Perplexity, Context7, and Firecrawl sources

Post-completion retrospective analysis with lessons learned

Multi-agent quality council review of completed work

Assess project risks with severity scoring and mitigation suggestions

Manage the file/command allowlist that blocks out-of-scope edits during the active task

ASCII progress dashboard for the active project

Bidirectional sync between local project state and integrated PM platform

Task CRUD operations: add, edit, complete, or block a task

Use or create reusable project templates (webapp, api-service, ml-pipeline, mobile-app, infrastructure)

Single execution cycle: research -> execute -> validate a single task

Builds and validates the task DAG. Detects cycles, computes critical path, returns the unblocked task set for the current execution cycle.

Extracts reusable patterns from completed tasks. Builds a pattern library for faster future decomposition and research. Runs after project completion.

Connects and syncs with 9 PM platforms (GitHub Projects, Linear, Notion, Asana, Trello, ClickUp, Monday.com, Todoist, Local). Credentials from CLAUDE_PLUGIN_OPTION_* env vars.

Tracks task completion counts and velocity. Picks the next task for execution. Updates progress/log.md. Lightweight — uses Haiku intentionally.

Conducts the 8-phase discovery interview for /pm:init. Asks one question at a time, never rushes, adapts to surprising answers. Requires minimum 15 questions before generating output.

Master coordinator for the autonomous project execution loop. Owns Phase 0-8 of /pm:auto, makes escalation decisions, manages HITL triggers, and coordinates all other agents.

Validates task completion against acceptance criteria. Each criterion is binary PASS or FAIL. Sets task COMPLETE or BLOCKED with specific evidence.

Decides whether deep research is needed for a task, checks the research cache, and dispatches to deep-researcher. Returns cached brief if fresh (< 24h).

Identifies unknowns and high-risk tasks. Creates spike tasks for unknowns. Flags tasks requiring HITL. Scores risk 1-10.

Converts interview output into a structured project definition with phases, milestones, success criteria, and risk register. Runs after project-interviewer completes.

Recursively decomposes phases into epics → stories → tasks → micro-tasks (5-15 min). Enforces INVEST principles and rejects tasks estimated over 30 minutes.

Implements one granular micro-task (5-15 min). Reads research brief first, implements atomically, saves artifacts. Refuses to execute without a research brief.

Serializes and restores project state. Writes phase checkpoints, enables resume from any checkpoint. Keeps rolling window of last 10 checkpoints.

Manages all state file reads and writes. Ensures session continuity across conversations. Handles atomic writes and corruption detection.

Multi-perspective review board for /pm:review. Evaluates completed work from quality, security, architecture, and UX angles simultaneously.

Performs deep pre-execution research for a task. Uses 4-source protocol: codebase grep first, then Perplexity/web search, then Context7 for library docs, then Firecrawl for specific URLs.

Connecting and syncing with GitHub Projects, Linear, Notion, Asana, Trello, ClickUp, Monday.com, Todoist, and Local

Rendering ASCII progress dashboards with phase/epic/task breakdown and velocity metrics

Decomposing projects into granular micro-tasks using 5-level hierarchy and INVEST principles

Using and creating project templates for webapp, API, ML pipeline, mobile, and infrastructure projects

Validating task completion against acceptance criteria with per-type automated checks

Deep research before task execution using 4-source protocol: codebase→Perplexity→Context7→Firecrawl

Managing project and task state in .claude/projects/{id}/ with atomic writes and session continuity

14 hooks across 6 events