Agent

Sensor

Adversarial code reviewer for the Luna AI Workforce. Reviews Pump's plans before code is written and Pump's PRs after. Read-only. Never approves when uncertain. Used by the platform team via Claude Code today; Slack DM wiring queued.

Live Platform team via Claude Code · Slack DM coming

← All agents

Sensor is the adversarial code reviewer on the Luna AI Workforce. Sensor reviews Pump’s scaffold plans before code is written, and Pump’s PRs after. Sensor is deliberately adversarial — the job is to find what Pump got wrong, not to be agreeable. Sensor never approves when uncertain.

Read-only, by design

Sensor never deploys, never pushes, never merges. The role is to surface findings — not to act on them. That’s why Sensor’s MCPs are read-only and the persona explicitly forbids any tool that mutates state.

Three review modes

Plan review — Pump submits a plan; Sensor returns APPROVED, NEEDS CHANGES, or BLOCKED with cited findings against the Cloudflare-Agents anti-pattern checklist (skills/shared/luna-agent-onboarding §9).
PR review — once Pump opens a PR, Sensor walks the diff, runs the §6 rubric, and posts findings as PR comments. PRs do not merge until Sensor’s findings are resolved.
Monthly fleet audit — runs on the 1st of every month against the deployed fleet. Output is a markdown report committed to Luna_AI_Workforce_Config/docs/sensor-audits/<date>-monthly.md plus one GitHub issue per ✗ finding.

Why this exists

Adversarial review at the start of a change is cheaper than catching the same problem after deploy. Sensor is the platform’s immune system.

Questions? john@lunadiabetes.com.

Try it

Chat with Sensor.

Same model and guardrails as Slack. Not saved to your conversation history — this is for quick trials.

Try: · ·