health

health

Read .agents/skills/health/SKILL.md and follow its workflow steps.

Important: rendering (the status table, non-clean sections, execution results) is handled by the health-check.mjs engine. Present the engine's stdout as-is; don't reformat or reinterpret it.

Claude Code-specific additions

Running the engine

Run health-check.mjs, in the same directory, via Bash (passing $ARGUMENTS as-is). In user-scope it's ~/.claude/skills/health/health-check.mjs; for dogfooding, <repo>/.claude/skills/health/health-check.mjs:

node ~/.claude/skills/health/health-check.mjs $ARGUMENTS

Argument parsing

Determine the presence of --deep / --fix / --yes. When --fix is absent (no arguments / --deep), the engine completes read-only, so present the output as-is and end. Any other flags are ignored (reserved for future extension).

--fix confirmation-gate wiring (follows the central policy)

When --fix is specified and --yes is absent, the engine resolves each action's gate via the central policy (policy-read.mjs). The host side branches the UI per gate (gate determination belongs to the engine — the host never re-determines the vocabulary/class):

1. Execute + list: run node ~/.claude/skills/health/health-check.mjs --fix [--deep]. The engine executes on the spot proceed (reversible-local) actions and outputs the "execution results" (✔ done / ✖ failed on each line) (audit trail, no confirmation needed). ask (irreversible stash drop) actions are output as "actions requiring confirmation (gate=ask)" and are not executed.
2. Individual confirmation of ask items: only when there's an ask list, confirm with AskUserQuestion per action whether it's OK to execute (don't enumerate options as text).
   • Approved → run node ~/.claude/skills/health/health-check.mjs --fix --yes [--deep] to execute the rest (the approved ask items), and present the execution results and the post-execution status table as-is.
   • Rejected → end without executing anything additional.
   • The ask list is empty (everything already executed via proceed) → present the execution results and post-execution status table, and end.

For --fix --yes (with --yes from the start), run the engine once, as-is, without inserting a confirmation. --yes is an explicit opt-out that overrides the policy, executing all safe actions, including those with gate ask (unattended execution via routine / /loop / schedule).

The engine's --fix safe boundary (executing only prune / delete / fetch / the --deep-promoted drop, with push / 要確認 / 要対応 / abort or continue out of scope) and gate resolution (reversible-local=proceed / irreversible=ask, falling back fail-safe to ask when a policy is absent) are enforced on the engine side, per the SKILL.md contract. The host side never re-determines the vocabulary or gate.

Completion report and next-action suggestions

End once the engine output has been presented. AskUserQuestion is not used other than for --fix's individual confirmation. No next-action suggestion is made either.