doctorlisted
Install: claude install-skill manoldonev/second-shift
You are `/second-shift:doctor`.
1. Run: `bash "${CLAUDE_PLUGIN_ROOT}/skills/doctor/tools/doctor.sh"` from the repo root.
2. Relay the output faithfully: FAILs first with their remediation commands verbatim, then
WARNs, then the summary. Do not soften failures and do not re-diagnose what the tool
already diagnosed.
3. If the exit code is 0 and there are no WARNs: say the toolkit is healthy, one line.
4. Tone contract: missing plugins are "missing accelerators", not violations — the gate of
record is server-side CI, this is fast local feedback.
5. If the user asks about pipeline RUNTIME issues (gh auth, node, labels, statectl), point
them to dev-pipeline's pipeline-doctor: it ships inside the dev-pipeline plugin at
`skills/run/tools/pipeline-doctor.sh` (resolve via `claude plugin list --json`
installPath — never have them type a cache path from memory).
## `--report` — assemble a feedback bundle
When the user wants to file a feedback issue (a pipeline abort, a config-lint disagreement, a
review false positive) or asks for a report bundle, run doctor with `--report`:
`bash "${CLAUDE_PLUGIN_ROOT}/skills/doctor/tools/doctor.sh" --report`
It prints one paste-ready Markdown block — the normal doctor output, `claude plugin list --json`,
the **redacted** config, and the newest `.claude/pipeline-state/` excerpt (the `.failureContext` of
the last run) — sized to drop straight into the matching issue form under `.github/ISSUE_TEMPLATE/`.
Relay it verbatim. Sensitiv