manage-claude-projects

原文涵盖的主题

• Manage Claude Code Projects
• Where the data lives (so you can explain it)
• Step 1 — Identify the project
• Step 2 — Show basic stats
• Step 3 — Offer actions
• Step 4a — Deep statistics
• Step 4b — Remove / clean up (destructive — be careful)
• Output style

Inventory, deep-stat, and safely clean up the projects Claude Code tracks under /.claude. All heavy lifting is done by scripts/projects.py (stdlib Python, prints JSON). Your job is to drive the interactive flow: pick a project, show stats, offer actions, and confirm anything destructive. Run everything as: python3 "${CLAUDEPLUGINROOT}/skills/manage-claude-projects/scripts/projects.py" A "project" exists in two places: /.claude.json → projects{ : {...}} — the registered project with its config, MCP servers, and last-session cost/token snapshot. /.claude/projects/ / — one .jsonl transcript per session. Encoded name = every non-alphanumeric char of the path replaced with -. Each session id (the .jsonl filename) also keys traces elsewhere: todos/ -, file-history/ /, and history.jsonl lines (matched by the project's absolute path). These are what removal targets. Resolve the target before doing anything else, in this order: 1. User gave a path or name — if they passed an absolute path, use it. If they gave a partial name (e.g. "staffing-assistant"), run list and match it against the project paths; if more than one matches, show the matches and ask which one. 2. No project given — run: T

这里只展示节选，请以完整原文和当前要求为准。