/fleet-doctor — Strict Fleet Health Check

Canonical surface: `scripts/coord/fleet-doctor-strict.sh` (INFRA-1427). Any harness invokes the same script; this skill is the Claude Code adapter.

Routing

Arguments passed: $ARGUMENTS. Optional flags: --json, --verbose.

scripts/coord/fleet-doctor-strict.sh $ARGUMENTS

The exit code IS the answer. Surface that explicitly to the user — don't bury it in prose.

What it checks (exit non-zero if any fail)

1. binary — chump binary exists and is not stale vs source
2. leases — no expired leases older than LEASE_STALE_HOURS (default 2h)
3. disk — free disk ≥ DISK_MIN_GB (default 5 GB)
4. dirty-prs — no open PRs in DIRTY state for > DIRTY_PR_HOURS (default 24h)
5. gap-drift — no unresolved gap-drift (open gaps with closed PRs, etc.)
6. p0-budget — open P0 gap count ≤ P0_MAX (default 5)
7. pillar-cover — every pillar has ≥ PILLAR_MIN (default 2) pickable gaps

When the user invokes this

• If exit 0: report "fleet healthy, all 7 checks pass" + summary of any borderline numbers
• If exit non-zero: list which checks failed, in order, and propose the next action (file a gap, rebuild binary, run a sweeper, etc.). The script's stdout has the diagnostic detail — surface it.

When NOT to use this

• For a snapshot-style summary, use /fleet-brief (no pass/fail; just status)
• For halt-class emergencies, use /operator-recall --check-only (different signal class)
• For PR-specific diagnostics, use gh pr view, pr-rescue.sh, etc.

Useful env overrides

LEASE_STALE_HOURS=4 DISK_MIN_GB=10 scripts/coord/fleet-doctor-strict.sh

Pass these via shell prefix when the user wants non-default thresholds.