health-check

name	health-check
description	Composite code quality score from linter, types, tests, and optional dead-code checks.

When To Use

• Weekly or pre-sprint baseline of repo health.
• After large merges or dependency upgrades.
• When the user asks "how healthy is the codebase?" or "run all checks."
• Before /retro to attach metrics to trends.

Related: /regression-check for deep test interpretation; /dependency-audit for supply chain; /skills-doctor for incident-driven "won't run" triage and deep debug (not baseline scoring).

Do not use this skill for security review or production monitoring substitution.

Core Stance

• Wrap existing project tools — do not invent custom linters.
• Score is weighted composite 0–10 with transparent breakdown.
• A failing gate is reported honestly — do not inflate scores.
• Track commands run and duration for reproducibility.

Research Backing

• Fowler (2006), Continuous Integration — automated quality gates on every change.
• Beyer et al. (Google, 2016), Site Reliability Engineering — measurable operational readiness.
• ISO/IEC 25010 — software quality model (maintainability, reliability characteristics).

Process

1. Discover toolchain
   • package.json scripts, Makefile, CI config, pyproject.toml, etc.
2. Run checks (when present)
   • Typecheck / compile
   • Linter
   • Unit tests (full or --changed if time-bound)
   • Optional: dead code (knip, ts-prune), shell lint
3. Score each dimension (0–10)
   • Types: pass=10, errors scale down
   • Lint: pass=10, warn/error counts reduce
   • Tests: pass rate and count
   • Build: success/fail
4. Weighted composite
   • Default weights: tests 40%, types 25%, lint 20%, build 15%.
   • If a dimension has no tool, exclude it and renormalize remaining weights to sum to 100%; note excluded dimensions explicitly.
5. Trend note
   • Compare to previous run if user provides prior score; else mark "baseline."

Operating Rules

• Run commands from repo root; note missing tools explicitly.
• Do not fix failures unless user requests — report only in this skill.
• Prefer same env as CI when documented (CI=true).
• Bun projects: use bun test, bun run per workspace rules.

Output Format

Return a markdown report with these exact sections:

• Toolchain Detected
• Commands Executed (with exit codes and duration)
• Dimension Scores (table: dimension, score, evidence)
• Composite Score (0–10)
• Top Failures (blocking)
• Recommended Fixes (priority order)
• Trend (baseline / delta if known)

Example

Dimension Scores

Dimension	Score	Evidence
Tests	8/10	142/145 pass, 3 flaky skipped
Types	10/10	tsc --noEmit clean
Lint	7/10	12 warnings

Composite Score

8.2 / 10