Saved to — applies immediately, no restart.
Every way teams put Foreman to work — each is copy-paste ready.
Turns on change tracking for every detected agent, MCP tracking, and the inbox — in one go. Run this first; nothing else to configure.
Runs the repo's own tests/build and attaches the real pass/fail, so an "unverified" claim becomes evidence — or is caught lying. Also the 🔬 button on any card.
Agents that earn a clean track record get their low-risk sessions auto-approved; anything risky still waits for you. Turn it on in Settings ⚙.
What agents actually pushed live — deploys, publishes, releases, pushes to main — flagging anything that reached prod without review.
Catches two agents editing the same file while both sessions were live — where last-writer-wins can silently drop the other's changes.
A portable, ed25519-signed, tamper-evident record of a session for any PR or release. Anyone verifies it offline. Also the 🔏 button on every card.
Auto-attest every MCP server already in your agents' configs (Claude/Cursor/Windsurf) — no URL-pasting. Undo any time with foreman unwire.
Claude Code, Cursor, Gemini CLI & OpenCode. Richest data: commands, claims, diffs — one command wires all four.
Codex, aider, anything that runs in a terminal — Foreman wraps the command and reviews what it changed.
Universal mode for Windsurf, Copilot, JetBrains AI — any tool that edits files. No hooks needed, just git.
Exits non-zero while unreviewed risky sessions exist — drop it into any pipeline or pre-push hook to block unreviewed AI work.
No install, no hooks, no journal — risk-scan the diff in CI and emit SARIF so findings annotate the PR itself. Just npx foremanjs.
Turn any card into a markdown evidence comment — what ran, what was claimed, what was proven. Also the 📋 button on every card.
Flag a card with a note — the agent reads it at the start of its next session in that repo and stops repeating the mistake.
Prefix any stdio MCP server: every tool call gets an ed25519-signed receipt, and tool-definition rug pulls land in your inbox.
Receipts are hash-linked — prove none was edited, deleted, or reordered since it was signed.
Share signed review packs through the repo you already share — teammates' sessions appear here with a 👥 badge, tamper-proof.
Live inbox counts and critical-card notifications on Windows, macOS (xbar/SwiftBar) and Linux.
Ignore paths, disable rules, change mass-rewrite thresholds, add a notify command — all in one config file.
Seed three example sessions (a destructive rewrite, a leaked secret, a clean verified fix) — remove them with --clear.
Full reference: foreman --help · docs in the README
Every AI agent session becomes a review card, ranked by risk. Work the inbox like email — approve what's safe, flag what isn't. Inbox zero means every agent change had human eyes on it.
— walks you through every option on this screen
Ignore paths, disable rules, or change thresholds in ~/.foreman/config.json — see foreman config.