Skip to content

Keesan12/Martin-Loop

v0.3.12 Breaking

This release includes 3 breaking changes for platform teams planning a safe upgrade.

Published 1mo AI Agents & Assistants
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-runtime ai-agent-runtime ai-coding-agents ai-control-plane ai-governance ai-safety
+13 more
audit-trail budget-enforcement claude-code codex coding-agents control-plane governed-runtime llmops mcp model-context-protocol observability policy-as-code rollback

Affected surfaces

auth rbac breaking_upgrade

Summary

AI summary

Governance is now mechanically enforced; martin gate and martin_run block without an estimate receipt.

Full changelog

The Fix

MartinLoop was promising governance but letting agents bypass it. This release makes governance mechanically enforced — not a suggestion.

What changed

martin estimate now writes a receipt

Running martin estimate "..." writes a timestamped receipt to workflow-state.json. This is the evidence trail that governance actually happened.

martin gate requires estimate

The gate now checks for both martin doctor AND martin estimate before returning exit 0. No estimate = BLOCKED. This is what the Claude Code PreToolUse hook calls — meaning Bash, Edit, and Write tool calls are blocked until governance is complete.

martin_run MCP tool blocks until estimate exists

The MCP server's run gate now includes estimate as a required step alongside doctor, plan, and preflight. Agents cannot call martin_run without first calling martin_estimate.

martin mcp install --host claude auto-installs hooks

Running martin mcp install --host claude now writes governance hooks directly into ~/.claude/settings.json. No manual copy-paste needed. The hooks install martin gate as a PreToolUse blocker for Bash/Edit/Write.

MartinLoop Memory Store

New _martin/memory.jsonl — append-only, never overwrites. Records user preferences, consent signals, budget patterns, and behavioral observations across sessions. Accessible via martin://agent/memory-summary MCP resource.

Install

npm install -g [email protected]

After install, wire governance into Claude Code:

martin mcp install --host claude

Breaking Changes

  • `martin gate` now requires a prior `martin estimate` receipt; without it the command exits non‑zero and blocks further tool calls.
  • `martin_run` MCP server run gate also mandates an existing `martin estimate` before allowing execution.
  • Removal of ability to bypass governance checks in previous versions.
View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track Keesan12/Martin-Loop

Get notified when new releases ship.

Sign up free

About Keesan12/Martin-Loop

All releases →

Related context

Earlier breaking changes

  • vmcp-v0.1.3 martin_status uses oneOf for selector exclusivity, latest as const.
  • vmcp-v0.1.3 maxIterations and maxTokens modeled as integers in tool schemas.
  • vmcp-v0.1.3 Tool schemas enforce additionalProperties: false on public contracts.
  • vmcp-v0.1.3 Packaged artifacts now require and ship server.json alongside package.json.

Beta — feedback welcome: [email protected]