martin-loop CLAUDE.md

Project instructions for Martin Loop, including its current implementation status and checks for governed AI runs. A governed run is an AI task executed with recorded controls and evidence.

In plain words
What is it for?
Use them when developing or testing Martin Loop’s CLI, verifying signed run records, checking build results, handling provider usage estimates, or reviewing feedback intake.
Why use it?
They tell contributors what already works, what has been fixed, and how to verify run receipts, builds, failures, pricing labels, and feedback delivery.

Instructions file

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/keesan12/martin-loop/claude-md
Clone the repo
git clone --depth 1 https://github.com/Keesan12/martin-loop
Per session 573 This file is loaded in full into every session.
When invoked 573 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00573 $0.00573
Opus 5 $0.00287 $0.00287
Sonnet 5 $0.00115 $0.00115
Haiku 4.5 $0.00057 $0.00057

Measured 2d ago against content hash 356370925495, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

martin-loop CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

CLAUDE.md · 48 lines

How it starts

The opening of the file, as written. The whole thing — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Claude Code Instructions

Current working status — 2026-08-11

Working now

  • Governed live runs work with both Codex (gpt-5.4) and Claude.
  • Successful runs persist signed receipts under ~/.martin/runs; martin runs verify --loop-id <id> --json verifies receipt integrity and martin dossier --loop-id <id> renders the evidence.
  • Verification commands are recorded with exit-code evidence. Real Claude and Codex runs against the OpenDesign test workspace passed pnpm build and produced verified receipts.
  • The CLI renders a verified completion state for successful governed runs and a signed-failure-receipt state for failed persisted runs.
  • Feedback and pilot-interest intake waits for the Supabase response and records a privacy-safe local delivery outcome: accepted, duplicate, queued, or rejected.

Fixed in the current fixes branch

  • OpenAI-compatible inferred usage or fallback model pricing is marked estimated, with estimatedUsd; only provider usage with known pricing is marked actual.
  • Windows Codex launches use the compatible --approve-for-me write mode and choose runnable npm shims.
  • Dirty worktrees are baselined before a governed attempt, preventing existing operator changes from becoming false scope violations; automatic rollback is skipped for a dirty starting workspace.
  • The stale Claude test expectation for Codex write mode was updated.

Still not complete

  • Intake delivery is recorded in local milestone state, not cryptographically bound to the already-finalized loop receipt.
  • Live agent output and full persisted agent transcripts are not yet implemented.
  • The remaining numbered findings in CTO_AUDIT_PLEASE.md / CTO_AUDIT.md are backlog unless individually marked fixed; do not describe the whole audit as resolved.

Read and obey AGENTS.md before making any change.

The most important repository rule is:

All MartinLoop implementation must be completed, tested, committed, reviewed, and merged in ML_Core_OSS_Internal before any public staging begins.

Read the full file on GitHub · 48 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 48 lines · 573 tokens per session scan A 356370925495

Subscribe to this mod's changes

martin-loop CLAUDE.md is an instructions file published in the GitHub repository Keesan12/martin-loop (47 stars, last pushed 2d ago), licensed Apache-2.0. It adds 573 tokens to every session, about $0.0029 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.