pm-skills

Project-management skills for a solo developer using AI coding agents. They sit above a development workflow skill set, such as Superpowers, and decide which feature is worked on. The workflow skills decide how it is built.

The skills are designed for literal, predictable, low-friction output: one next action, an explicit "done when", and fixed templates.

• Parking lot: ideas, bugs, questions, and inspiration are captured in one line at any time, even mid-task, by you or the agent, and discussed later.
• Bugs with the facts a fix needs: pm-bug asks for the symptom, expected and actual, reproduction, severity, and evidence, checks for a duplicate, and routes the defect to a fix or the roadmap.
• A risk register with RAG status: pm-risk captures a risk in one line at any time, even mid-task, then adds, updates, re-rates, reviews, closes, or removes it in planning/RISKS.md. Each risk's red, amber, or green status is derived from its likelihood and impact, with a recorded override, and a mitigation can become a roadmap item.
• Corrections, not pretences: pm-revise handles built work that is not what you wanted. It rewinds the active item to an earlier stage, fixes it in place, or opens a revision item that revisits the original. It is the one PM skill allowed to interrupt an active workflow, and it writes the correction down first.
• Brainstorming with challenge: pm-shape interviews, challenges, and maps the ramifications of an idea before it reaches the roadmap. pm-discover analyses the existing code, history, and usage for evidence. Every idea, including the agent's own, is challenged.
• Strategy, roadmap, and board: pm-strategy sets milestones, bets, and focus, and generates the roadmap. pm-roadmap renders the full milestone roadmap, pm-board the task board, pm-up-next the ordered queue, and pm-status the summary and your own queue. Move, add, and insert items from the board or roadmap with the shared ordering rules. Items are owned by the agent, the user, or both.
• The plan as data: pm-export writes every roadmap item, board column, milestone, bug, brief, risk, and log entry to planning/planning.json in one fixed shape, with the JSON Schema that describes it, or checks the file without writing. The Markdown stays the source of truth.
• The plan as a site: pm-docs-site plans a local Astro and Starlight site that renders planning.json as Home, Board, Roadmap, Log, and Risks pages, then builds it from shipped templates inside that item's workflow. Optional modules add the repository's docs/ pages, LikeC4 architecture views, and an Aspire AppHost resource. It reads the export only.
• Purview fit: for .NET projects, pm-purview recommends Purview tools only where they remove real work. For an empty project it asks first what you are building and at what scale. Non-.NET projects get no suggestions.
• Memory across sessions and compaction: every decision is written to a file the moment it is made. pm-resume restores the state afterwards. pm-rejoin puts a session back on the lifecycle after the thread is lost.
• Commits without the chore: the skill that writes the planning files commits them, on the branch you plan on; a completion record is committed on the item's branch so it rides the item's pull request. The default is automatic, set in config.md (ask or never instead), and the same commit refreshes planning.json when it exists.
• Questions use the agent's question UI: choices with a recommended option and free text. You are never asked to type yes or a/b/c.

Install

Run this from the root of the target repository:

npx skills add https://github.com/purview-dev/pm-skills -y
# or
bunx skills add https://github.com/purview-dev/pm-skills -y

Skills install at project level by default. Add -g for a user-level install, or -a claude-code to target a single agent. See the skills CLI for all options.

The skills hand design, planning, and execution to a workflow skill set, by default Superpowers. pm-setup checks that every workflow skill is installed and prints the command when one is missing:

npx skills add https://github.com/obra/superpowers -y
# or, in Claude Code
/plugin install superpowers@claude-plugins-official

Then, in your agent:

1. run pm-setup merges the rules into your existing AGENTS.md or CLAUDE.md, between <!-- pm-skills:start --> and <!-- pm-skills:end --> markers. Nothing outside the markers changes.
2. Commit, using the command pm-setup prints.
3. run pm-init writes planning/VISION.md by default and seeds the roadmap. pm-setup asks where to store the planning files before its other setup questions; choose a different repository-relative directory if needed.
4. Not sure what to run? pm-help explains every skill, lists what is missing in your setup, and recommends the next skill.

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗