Keesan12/martin-loop

View on GitHub ↗

Make AI coding agents safe to scale autonomously: assign work, cap spend, enforce policy, verify output, roll back failures, learn from loops, and prove ROI across every repo.

44 ★3 forksTypeScriptUpdated 18d ago

What you need to know

Governed execution layer for AI coding agents (Claude, Codex, Gemini, Cursor) that adds budgets, verifier gates, scope, and audit receipts to every run.

Install

npx -y martin-loop@latest start

Key features

  • ✓Budget caps (USD/token/iteration), soft limits, and stop conditions
  • ✓Verifier gates (e.g. npm test) before a run counts as complete
  • ✓13-class runtime failure taxonomy
  • ✓Proof receipts: run-receipt.json/md with integrity verification
  • ✓CLI, SDK, and standalone MCP package (@martinloop/mcp)
  • ✓Adapter support for Claude, Codex, Gemini, OpenAI-compatible providers
  • ✓Benchmark workspaces and GitHub Actions budget gate example

Best for

Teams that need bounded, inspectable, and auditable AI coding runs.

Platforms: Node.js 20+

Reviewed 2026-08-11

Topics

agent-runtimeai-agent-runtimeai-coding-agentsai-control-planeai-governanceai-safetyaudit-trailbudget-enforcementclaude-codecodexcoding-agentscontrol-planegithub-actionsgoverned-runtimellmopsmcpmodel-context-protocolobservabilitypolicy-as-coderollback
Stars
44★
Forks
3
Language
TypeScript
License
Apache-2.0
Created
2026-04-04
Last push
2026-08-17