Profile

Autopilot dashboard

Every number MUST be traceable to a real entry in the data files. Never fabricate or extrapolate a value claim. If a number is zero, show zero.

Data sources

Read from ~/.claude/command-autopilot/: state.json (counters, milestones, config), events.jsonl + events-archive.jsonl (the ledger), learned.json (rules), skills-index.json.

Report (user's language, compact)

1. Value ledger — counts from events: auto plan-mode entries, skills auto-invoked (top 3 by count, with names), suggestions accepted vs made. One line each, with the period covered.
2. Before/after — compare the oldest 7 days of events against the newest 7 (only if ≥14 days of data; otherwise say "still collecting"): self-used commands per week then vs now. This is the user's own growth, shown from their own data.
3. Learned rules — each in_force rule with its evidence count and last-confirmed date. If none: "still observing, first rules typically appear after ~10 sessions."
4. Mastered — habits with self-use evidence; note that teaching for them has naturally stopped.
5. Controls — one line: config skill to quiet/mute, evolve skill to re-distill, "delete rule N" works right here.

Feedback draft

If the user is dissatisfied with any suggestion behavior, offer to draft a GitHub issue for WinterDDo/claude-code-skill-autopilot: title, what happened, expected behavior, and (only with explicit consent) the relevant anonymized evidence lines. Show the draft; the user files it themselves.

Rule deletion

"Delete rule N" → remove it from learned.json, confirm in one line.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: WinterDDo
• Source: WinterDDo/claude-code-skill-autopilot
• License: MIT

Install and usage instructions live in the source repository linked above.