claude-task-skills
Portable task governance skills for Claude Code. Adds structured, folder-based task tracking to any project — create tasks, close them cleanly, audit open work, and manage session lifecycle with consistent governance.
These skills are project-agnostic. They discover your project's conventions at runtime by reading CLAUDE.md, scanning for task trackers, and adapting. They work out of the box with zero configuration, and layer on top of whatever project-specific behaviors you define in your own CLAUDE.md.
Installation
All /plugin commands below are run inside a Claude Code session (CLI, desktop app, or IDE extension). They are Claude Code slash commands, not terminal commands.
From GitHub
Open Claude Code and run:
/plugin marketplace add avanrossum/claude-task-skills
/plugin install claude-task-skills@claude-task-skills
The @claude-task-skills suffix is the marketplace name (set in .claude-plugin/marketplace.json). The plugin and the marketplace happen to share that name here, which looks repetitive but is correct.
From a local clone
Clone the repo, then start Claude Code with the plugin loaded:
git clone https://github.com/avanrossum/claude-task-skills.git
claude --plugin-dir ./claude-task-skills
For development / testing
claude --plugin-dir /path/to/claude-task-skills
Reload after making changes (inside a Claude Code session):
/reload-plugins
Skills
/create-task-env — Bootstrap Task Governance
One-time setup for projects that don't have task infrastructure yet. Run this first — then all the other task skills work automatically.
/create-task-env
What it creates (skipping anything that already exists):
tasks/ directory for task folders
TASKS.md task tracker at the project root
CHANGELOG.md for change history
- Task conventions section appended to
CLAUDE.md (or creates one)
- Optionally:
ROADMAP.md and LESSONS_LEARNED.md (asks first)
Non-destructive — safe to run on a project that already has partial governance. It fills gaps without overwriting.
/create-task — Start a New Task
Creates a task folder, stages governance files, updates the task tracker, and — critically — verifies that any existing in-progress tasks have clean governance before opening a new one.
/create-task build a CSV export feature for the dashboard
/create-task start the blast radius analysis from the roadmap
What it does:
- Parses your description (or looks up a reference from your roadmap/backlog)
- Checks all in-progress tasks for governance hygiene — fixes stale docs before proceeding
- Discovers your project's task conventions from
CLAUDE.md (or proposes sensible defaults)
- Creates the task folder and stages a TASK.md with title, summary, spec, and status
- Adds a "task open" entry to your task tracker (
TASKS.md or equivalent)
/close-task — Close a Task
Finalizes a task and updates all governance surfaces. Project-specific close actions (change control docs, Asana posts, Notion updates, etc.) are defined in your CLAUDE.md — the skill handles the universal steps.
/close-task
/close-task regional rep goals dashboard
Standard actions (always happen):
- Finalizes the task document (status, changes made, spec accuracy)
- Updates the changelog
- Moves the task to "Completed" in the task tracker
- Updates the roadmap (checks off completed items)
- Runs any project-specific close procedure from
CLAUDE.md
/list-tasks — List Tasks by Status
Quick status check across all tasks. Filters by status, cross-references the tracker against actual task folders, and flags discrepancies.
/list-tasks # Open tasks + completed count
/list-tasks open # Only in-progress tasks
/list-tasks completed # Only completed tasks
/list-tasks stalled # Blocked or stalled tasks
/list-tasks all # Everything
/task-audit — Progressive Task Audit
A collaborative, one-at-a-time walk-through of all open tasks. The agent does a silent governance pre-pass first, then walks you through each task that needs attention — surfacing stale status, missing docs, scope creep, and unresolved blockers.
/task-audit
How it works:
- Silent scan of all open tasks, noting governance issues
- Opening summary: "Found N open tasks. X healthy, Y need attention, Z may be stale."
- Progressive walk-through — one task at a time, starting with problems. Waits for your input before moving on.
- Quick confirmation pass on healthy tasks
- Wrap-up with summary of changes made
How It Works
These skills are designed around a simple principle: CLAUDE.md is the extension point.