health
by tw93health runs a budget-aware Agent Health audit for Codex, Claude Code, Pi, agent instructions, hooks/MCP, verifier surfaces, and AI maintainability. Use the health skill to check why an agent ignores instructions, misses validation, or drifts into hard-to-maintain behavior. It is especially useful for Security Audit workflows, but not for debugging code or reviewing PRs.
This skill scores 74/100, which means it is listable and useful for directory users, but still somewhat limited. The repository provides a real, triggerable health-audit workflow for agent configuration and AI maintainability, so installing it should reduce guesswork versus a generic prompt; however, users should expect some rough edges from placeholder markers and the need to interpret a fairly large, script-backed skill body.
- Strong triggerability: the frontmatter names concrete use cases like checking Claude/Codex/Pi configs, ignored instructions, missing validation, and maintainability drift.
- Substantial operational detail: the skill body is large, has multiple headings, code fences, and explicit audit flow across agent config, instruction surfaces, tools/runtime, verifiers, and maintainability.
- Good agent leverage: includes an install command plus supporting scripts for context, doc-reference, maintainability, and verifier-output checks, which suggests repeatable workflow support.
- Placeholder markers ('todo', 'placeholder') are present, which lowers trust that every part of the workflow is fully polished.
- No references/resources/rules support files were found, so users may need to rely primarily on the main SKILL.md and scripts for adoption.
Overview of health skill
What the health skill does
health runs a budget-aware agent health audit for Codex, Claude Code, Pi, agent instructions, hooks/MCP, verifier surfaces, and AI maintainability. It is for readers who need a practical health guide for understanding why an agent ignores instructions, skips verification, or drifts into hard-to-maintain behavior.
Best fit and main job
Use this health skill when you need a fast configuration and workflow check, not a code review. It is especially useful for Security Audit-style work because it surfaces whether the agent stack is trustworthy enough to rely on before you inspect app logic, secrets handling, or policy compliance.
What makes it different
The skill is opinionated about layered agent health: config surfaces, instruction files, tools/runtime, verifiers, and maintainability. That means the result is more decision-ready than a generic prompt because it can point to the misaligned layer instead of just saying “something looks off.”
How to Use health skill
Install context and first files
Install the health skill with npx skills add tw93/Waza --skill health. After install, read SKILL.md first, then inspect agents/inspector-context.md, agents/inspector-control.md, and agents/inspector-maintainability.md to understand what the audit expects and what evidence it prioritizes.
What input the skill needs
The health skill works best when you provide repo-specific signals: agent instruction files, hooks or MCP settings, verifier commands, and any symptoms like “Claude ignores AGENTS.md” or “Codex config looks right but behavior is inconsistent.” If you want a useful health usage result, include the exact repo path, the agent platform, and the failure mode you saw.
How to phrase a strong request
A weak request is “check health.” A stronger one is: “Run a health audit on this repo for Claude Code and Codex. Focus on instruction layering, hooks, verifier coverage, and whether AI maintainability is likely to degrade. Call out any broken paths, stale references, or missing verification surfaces.” This gives the skill enough scope to produce a decision-useful audit.
Workflow and repo-reading order
Use the repository’s collection scripts first if available, then inspect the audit outputs before reading raw source. In this repo, the practical paths are scripts/check_agent_context.py, scripts/check_doc_refs.py, scripts/check_maintainability.py, and scripts/check_verifier_output.py. That order helps you identify where the problem is before you manually inspect every file.
health skill FAQ
Is this only for setup checks?
No. The health skill is for agent configuration audits, instruction drift, verification gaps, and maintainability risk. It is not meant for debugging application bugs or reviewing a pull request diff.
When should I not use the health skill?
Do not use it if you only want a code-level fix, a bug diagnosis, or a one-off prompt rewrite. If the repo is tiny and has no agent surfaces, a plain prompt is usually enough; the health skill is most valuable when there are multiple agent files, hooks, or verifier layers.
Is the health skill beginner-friendly?
Yes, if you can provide the repo and the symptom. The main beginner mistake is asking for a vague “health check” without saying which agent, which instruction surface, or which verification step seems broken. The more precise the input, the more actionable the output.
How does it compare to a generic prompt?
A generic prompt can summarize concerns, but the health skill is built to inspect agent surfaces systematically and budget its attention. That matters when you need a health install decision for Security Audit workflows, where missing validation or stale instructions can quietly undermine trust.
How to Improve health skill
Give it the right symptom and boundary
State whether you care about Claude Code, Codex, Pi, or all three, and name the failure: ignored instructions, missing hooks, weak verifier coverage, or AI maintainability drift. If you know the boundary, the skill can avoid wasting attention on unrelated repo structure.
Provide evidence, not just a conclusion
Stronger inputs include the relevant files or excerpts: AGENTS.md, local instruction overlays, settings, hook config, and any command output that shows the failure. For example, “AGENTS.md says verify with make test, but the repo only has pytest -q” is much better than “tests seem off.”
Iterate after the first audit
Use the first result to decide whether the problem is instruction layering, tool/runtime mismatch, or verifier weakness, then re-run with a narrower target. If the skill says the agent stack is healthy but maintainability is weak, ask for a second pass focused on TODO density, hotspot ownership, and stale doc references instead of re-auditing everything.