vigiles

vigiles

Make the harness your AI agent runs on reliable.

---

Agent = Model + Harness. The model gets the headlines — but the harness (your CLAUDE.md, hooks, and skills) is the half you control, and it fails silently: a stale reference, a guard that fails open, a skill that never fires. vigiles[^name] makes it prove itself — verify, gate, test, measure — mostly free, no API key; the real-model checks run on your Claude Pro/Max subscription, not metered tokens. Four instruments, adopt any:

🔎 Lint	Your CLAUDE.md stops lying to the agent — every path, script, symbol & linter rule checked against reality, not just well-formed. →
🛡 Guard	A safety hook that can't silently fail open — write it as a typed function, vigiles compiles a guard that actually blocks. →
🧪 Test	Prove your hooks, skills & subagents actually do their job — free, no API key. →
📊 Eval	Know if a skill helps or just costs — A/B on real tasks: the bill + correctness, on your own subscription. →

Pick the one that hurts today. No TypeScript? vigiles lint + inline <!-- vigiles:enforce --> comments work on any CLAUDE.md. Ship plugins? vigiles scan ranks a marketplace by structural health, no key. Works with Claude Code and Codex (vigiles/codex), or teach it your own harness.

Quick start

Paste into Claude Code or Codex:

Set up vigiles in this repo with good defaults (lint + test, non-interactive).
Verify my CLAUDE.md / AGENTS.md references and show me what's stale, then write
and run a harness test for one of my hooks or skills. Ask me first before gating
it in CI, adding a real-model eval, or enforcing strictly (--strict).

Or do it yourself:

npx vigiles init   # sets up lint + test: spec + harness test + CI + plugin

Interactive in a terminal, non-interactive for agents/CI (or --yes).

You don't hand-write any of this — your agent does. init installs model-invocable skills, so a plain-English ask does the work (it edits the source and recompiles on save; you never touch it by hand):