v0.1.1 — Claude Code plugin format

Adds .claude-plugin/plugin.json + hooks/hooks.json so the repo is installable via the official Anthropic plugin marketplace at https://claude.ai/settings/plugins/submit. Standalone curl install path unchanged.

v0.1.0 — initial release

First public cut.

A Claude Code Stop hook that pattern-matches positive-closeout vocabulary in the assistant's outgoing message and blocks turns that lack matching evidence. Returns a repair-guidance template the model self-corrects from on the next turn.

What's inside

• no-vibes.sh — the hook (370 lines bash, jq-only dep)
• settings.example.json — drop-in Claude Code wiring
• README.md — installation + scope
• RECEIPTS.md — six reproducible local tests + four redacted real fires against Claude Opus 4.7

Hook event coverage

• Stop and SubagentStop — false-success and missing-evidence checks
• PreToolUse(Bash) — destructive command patterns require explicit human approval
• PreToolUse(Write|Edit|MultiEdit) — refuse writes to .env, .env.*, .claude/*.local.json, secrets/**
• TaskCreated — implementation-like tasks require explicit ownership
• TaskCompleted — subagent closeouts get the same false-success / missing-evidence checks

License

Apache-2.0