pi-warden
Guardrails for Pi built on pi-typesafe that steer the agent instead of interrupting you: Jev judges irreversible and off-task tool calls, detects stuck
What it does
Guardrails for Pi built on pi-typesafe that steer the agent instead of interrupting you: Jev judges irreversible and off-task tool calls, detects stuck loops, checks unverified done claims, flags slop. Agent plumbing where probabilities beat regex on model output.
Fork the pattern, not the prose. Tags in the wild: agent tooling, guardrails, agents. The repo has install steps, env vars, and recent commits if you are deciding whether to depend on it.
How Jev is used
Multiple guards call Jev with context about the user request, tool args, and project rules. Verdicts quote violated Markdown rules, flag irreversible commands, or detect runaway output.
Setup: pi install npm:pi-warden. Add pi-warden.md at repo root with one heading per rule, or rely on README/CLAUDE.md/AGENTS.md fallback.
Steer instead of stop
Warden's default is to tell Pi what went wrong so it can fix the issue and continue. That keeps long tasks flowing while still enforcing security and style constraints Jev can judge from text.
FAQ
- How is pi-warden different from pi-jev-auto-mode?
Auto-mode focuses on approving individual tool calls. Warden supervises broader failure modes: rule violations on writes, stuck loops, dishonest done claims, and security patterns.
- Where do rules live?
pi-warden.md headings become individual rules Jev judges on every write and edit. See docs/configuration.md in the repository.
- Does it need a TypeSafe key?
Jev-powered guards need API access. Some offline pattern guards work without keys per README.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.