karlorz/agent-skills/harness-doctor

harness-doctor

Read-only audit. Do not disable, uninstall, delete, edit settings, or widen permissions.

Run

From this plugin (repo checkout or installed copy):

python3 scripts/harness-doctor.py --harness all
python3 scripts/harness-doctor.py --harness claude|grok|codex|cursor
python3 scripts/harness-doctor.py --harness all --json --days 30 --limit 50

Installed path is next to this skill: ../../scripts/harness-doctor.py. --home is for fixtures only.

Modes

• /harness-doctor or /harness-doctor all: every detected harness.
• /harness-doctor claude|grok|codex|cursor: one harness.
• Print the human report unless the user asked for JSON.

Rules

• Missing signals are UNKNOWN, never a guessed zero.
• Passive components (theme/output/monitor/LSP/obsidian) are not cold from zero.
• Maintenance tools (cursor-github-marketplace-repin, grok-build-harness, host-backup-restore) are rare_keep.
• Fewer than 20 sessions or under 7 days: no retirement verdict.
• Do not launch Grok /learn. If the collector exists, say it is the qualitative follow-up.
• Never print secrets, prompt bodies, account ids, or token-bearing URLs.
• Every disable/delete note is reversible and needs a separate attended step.

After the report

Stop. Do not apply cleanup unless the user explicitly asks in a later turn.