# Doctor

> Diagnose whether Skill Autopilot's hooks are actually firing and config is valid. Use when the autopilot seems inactive, suggestions never appear, after install or update to verify setup, or the user says "autopilot不工作", "check autopilot", "autopilot broken".

- **Type:** Skill
- **Install:** `agentstack add skill-winterddo-claude-code-skill-autopilot-doctor`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [WinterDDo](https://agentstack.voostack.com/s/winterddo)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [WinterDDo](https://github.com/WinterDDo)
- **Source:** https://github.com/WinterDDo/claude-code-skill-autopilot/tree/main/plugins/skill-autopilot/skills/doctor

## Install

```sh
agentstack add skill-winterddo-claude-code-skill-autopilot-doctor
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Autopilot health check

Two independent checks. Report both, then a one-line verdict.

## Check 1 — self report (only the model can confirm this)
State plainly, first: did the CURRENT prompt's context contain a block starting with `[AUTOPILOT]`? Yes or no. If yes, injection reaches the model — the core product works.

## Check 2 — the canary
Run:

```
sh "${CLAUDE_PLUGIN_ROOT}/hooks/run.sh" doctor-noop 2>/dev/null; python3 "${CLAUDE_PLUGIN_ROOT}/scripts/doctor.py" "$HOME/.claude/command-autopilot" ""
```

Read the output. Key line is `router canary`: if it shows THIS session, the hook fired end-to-end. If it says NEVER FIRED, tell the user to restart Claude Code (hooks load at startup) and run this skill again.

## Check 3 — skills surfacing (read-only; the Skills Autopilot layer)
Report what the skill layer sees and has done. All read-only, no writes:

```
python3 - <<'EOF'
import json, os
d = os.path.expanduser("~/.claude/command-autopilot")
try:
    idx = json.load(open(d + "/skills-index.json")); b = idx.get("budget", {})
    print("skills indexed:", len(idx.get("skills", [])),
          "| parked (native can't auto-fire these):", b.get("parked_count"),
          "| over native budget:", b.get("over_budget"))
except Exception:
    print("no skills index yet — restart Claude Code so SessionStart builds it")
try:
    s = json.load(open(d + "/state.json")); w = s.get("counters", {}).get("wake", {})
    f = lambda k: sum(v.get(k, 0) for v in w.values())
    print("wake popups — shown:", f("shown"), "accepted:", f("accepted"), "declined:", f("declined"))
except Exception:
    pass
EOF
```

Report in one line: N skills installed, M parked (invisible to native — the ones the autopilot uniquely surfaces), and the wake shown/accepted tally. If `shown` is still 0 after real use, surfacing isn't firing yet — say so honestly; the real ledger is the judge, not assumptions.

## Verdict
- Both pass → "Autopilot is healthy." Show the config line so the user sees their current mode.
- Check 1 fails but files look fine → restart needed, or another hook is crowding the context; suggest restart first.
- Python missing in check 2 → explain stateless mode honestly: core rules still work, learning and evidence are paused until Python 3.8+ is installed.

## Source & license

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

- **Author:** [WinterDDo](https://github.com/WinterDDo)
- **Source:** [WinterDDo/claude-code-skill-autopilot](https://github.com/WinterDDo/claude-code-skill-autopilot)
- **License:** MIT

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

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** yes
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-winterddo-claude-code-skill-autopilot-doctor
- Seller: https://agentstack.voostack.com/s/winterddo
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.