Night Market Diagnostics Toolkit

Interpretation notes

• context_optimizer stats prints nothing in the default text format. This is a verified quirk: stats on a directory with no --format flag produces empty output and exit 0. Always pass --format json (clean JSON) rather than --format table, which dumps a raw Python dict.
• Ratchet failures name the new violations. A ratchet that fails after your change means your change added debt. Fix the new ref or add the missing section. Never raise a baseline number to make a check pass, with one documented exception: a brand-new library skill legitimately starts uncalled, and scripts/check_skill_graph_drift.py itself (plus the _comment in scripts/skill_graph_baseline.json) instructs you to raise max_uncalled_libraries to record the 30-day consumer grace period that .claude/rules/shared-utility-consumer-rule.md grants (the rule grants the grace period; the script defines the baseline mechanics). night-market-architecture-contract records the same exception in its ratchet invariants.
• Mutation exit codes are the contract. CI (.github/workflows/mutation-testing.yml) treats exit 0 and exit 2 as success and posts survivors to the step summary. Only exits other than 0 and 2 fail the workflow. So a green mutation run does NOT mean zero survivors. Read the report.
• check_pinned_versions needs network access. It queries GitHub for latest releases. Offline, expect failures that are environmental, not real findings.

Health snapshot in one command

scripts/health-snapshot.sh (in this skill directory) runs the six cheap checks and prints a PASS/FAIL table. It is read-only, needs only bash and python3, requires no network, and completes in about 6 seconds (measured 2026-07-02).

bash .claude/skills/night-market-diagnostics-toolkit/scripts/health-snapshot.sh

It runs: plugin structure validation (all plugins), capabilities sync, supply-chain scan, both drift ratchets, and the description budget. Exit 0 means all six passed. Exit 1 means at least one failed, and the failing check's full output is printed to stderr above the table. Run it at session start, before a PR, and after any bulk skill edit.

Blocking vs advisory

Which failures stop a merge and which are signals:

The dimension-level policy lives in .claude/quality_gates.json: enforce_blocking is true with max_critical_issues: 3, and only the security dimension sets block_on_violation: true. The performance (20KB / 5,000-token file limits, 60-line functions, complexity under 12), maintainability (debt ratio under 0.3, nesting under 5), and compliance dimensions are advisory (block_on_violation: false). So: security findings block, size/complexity findings warn.

When NOT to use

• Running the test suite, lint, typecheck, or a release: use night-market-operations (command anatomy and artifacts).
• You already have a failure and need symptom-to-cause triage: use night-market-debugging-playbook.
• You need the meaning and default of a config knob rather than a measurement: use night-market-config-catalog.
• You are deciding whether evidence is sufficient to claim work complete: use night-market-validation-and-qa.

Exit Criteria

• bash .claude/skills/night-market-diagnostics-toolkit/scripts/health-snapshot.sh runs from the repo root, prints a six-row PASS/FAIL table, and exits 0 when all checks pass and 1 when any fails.
• Every invocation in the tool table is copy-pasteable from the repo root and executes without a usage error.
• A reader who has only this file can state the ratchet contract: baseline freezes existing debt, new violations fail, shrinking the baseline locks in wins.
• The blocking/advisory table matches .pre-commit-config.yaml entries and the block_on_violation flags in .claude/quality_gates.json.
• Mutation exit-code semantics (0 clean, 2 survivors tolerated, anything else is a crash) match .github/workflows/mutation-testing.yml.

Provenance and maintenance

Compiled 2026-07-02 against repo v1.9.15 on branch discussions-fix-1.9.14. Volatile numbers observed that day: 197 skills, 134 commands, 56 agents registered; description budget 73,135 of 90,000 chars used (81.3%); ratchet live counts 5 dangling refs (baseline 31), 7 uncalled libraries (baseline 8), 1 missing Exit Criteria (baseline 127); 13 version pins current with bandit intentionally held at 1.8.6.

Re-verification one-liners:

# Baselines and live ratchet counts
python3 scripts/check_skill_graph_drift.py; python3 scripts/check_skill_exit_criteria_drift.py

# Registration counts and budget usage
bash scripts/capabilities-sync-check.sh | tail -3; python3 plugins/abstract/scripts/validate_budget.py | tail -3

# Pre-commit wiring of the blocking table
rg -n "check_skill_graph_drift|check_noqa|validate_budget|check_pinned_versions" .pre-commit-config.yaml

# Mutation exit-code contract
rg -n "exit_code" .github/workflows/mutation-testing.yml

# Gate dimension block flags
python3 -c "import json;print(json.load(open('.claude/quality_gates.json')))"