\n \n\n **Trust your coding agents with real work.**\n\n Built from thousands of agent runs where the problem was not just intelligence — it was whether you could give the agent a real job and confidently walk away.\n\n **Measure your loop tax:** `npx -y martin-loop@latest audit` \n **Get started:** `npx -y martin-loop@latest start` \n **Try the demo:** `npx -y martin-loop@latest demo`\n\n [](./LICENSE)\n [](./tsconfig.base.json)\n [](#quick-start)\n [](https://www.npmjs.com/package/martin-loop)\n [](https://www.npmjs.com/package/martin-loop)\n\n MartinLoop is part of the NVIDIA Inception program.\n \n \n \n \n \n
\n\n## Start Here\n\n**Audit first** — run `npx -y martin-loop@latest audit` to see how much API-equivalent Claude Code spend went into fix-and-retry loops. Add `--offline` for zero network access or `--share` for a shareable summary.\n\n**Install / onboard** — run `npx -y martin-loop@latest start`, or install it globally with `npm install -g martin-loop@latest`.\n\n**Governed run** — define an objective, verifier, budget, and iteration cap with `martin run`.\n\n**Verifier** — completion requires fresh verifier evidence bound to the active run and workspace. A configured verifier proves only the checks it runs; `VERIFIED` is not a claim that the code is bug-free or automatically safe to merge.\n\n**Budget** — set a hard spend ceiling with `--budget-usd` and an attempt ceiling with `--max-iterations`.\n\n**Receipts** — inspect the latest result with `martin dossier --latest` and validate stored integrity with `martin runs verify --latest`.\n\n**Hosted sync (optional)** — governed work is local-first. Configure `MARTIN_API_TOKEN` and `MARTIN_TELEMETRY_ENDPOINT`, then use `martin sync status` and `martin sync flush` to send preserved evidence to a dashboard later. See the [quickstart](./docs/getting-started/quickstart.md#optional-hosted-dashboard-sync).\n\n**MCP** — install `@martinloop/mcp@latest` in a supported host or generate host configuration with `martin mcp print-config`.\n\n**Documentation** — continue with the [quickstart](./docs/getting-started/quickstart.md), [CLI reference](./docs/reference/cli.md), or [MCP setup](./docs/getting-started/mcp.md).\n\nWhen `--model` is provided, MartinLoop passes it through unchanged. Without `--model`, the authenticated host runtime chooses its own default. MartinLoop does not inject a hidden fallback model.\n\n## For AI Agents\n\nMartinLoop is the system around the coding job.\n\nThe coding agent still writes the code. MartinLoop keeps the job, limits, verification, recovery, outcome, and history consistent around the work.\n\nUse MartinLoop when a coding task needs:\n\n- clear job and success criteria\n- preflight readiness before agent spend\n- hard budget and iteration limits\n- allowed or denied edit scope\n- controlled retries and stop conditions\n- independent verifier commands\n- recovery and rollback evidence\n- a finite completion decision\n- run receipts, dossiers, history, failure classification, and post-run analysis\n- one consistent workflow across supported coding-agent runtimes\n\nCanonical lifecycle:\n\n```text\nDEFINE\n -> PREFLIGHT\n -> CONTROL\n -> VERIFY\n -> RECOVER\n -> PROVE\n -> ANALYZE\n```\n\nThe product-level flow is **Definition of Done -> Controlled Run -> Verified Handoff**.\n\nFor machine-readable context start with [`llms.txt`](./llms.txt), [`llms-full.txt`](./llms-full.txt), and [MartinLoop for AI Agents](./docs/for-agents.md).\n\n## One System Around the Run\n\nTeams should not need to stitch together a separate script or point tool for every part of coding-agent execution. MartinLoop connects the control path around the agent from preflight through post-run evidence.\n\n| Stage | MartinLoop role |\n| --- | --- |\n| Define | Capture the objective, verifier, budget, scope, and finish line. |\n| Preflight | Check readiness and required workflow evidence before agent spend. |\n| Control | Enforce budgets, attempts, path boundaries, policy, and stop conditions while the coding agent works. |\n| Verify | Run configured checks and bind the evidence to the active run and workspace. |\n| Recover | Preserve recovery and rollback state when another attempt or human review is required. |\n| Prove | Produce the authoritative `VERIFIED`, `STOPPED`, or `NEEDS REVIEW` handoff plus receipts. |\n| Analyze | Inspect run history, cost provenance, failure classes, dossiers, and shareable evidence after execution. |\n\nMartinLoop does not replace Git, GitHub, CI, dedicated security scanners, observability platforms, code review, or the coding agent itself. It gives those workflows one governed execution record to inspect.\n\n### Change the worker. Keep the job.\n\nMartinLoop is model-agnostic by design. The job, limits, verification, recovery, and outcome are separate from the worker underneath.\n\nNative coding-agent CLIs and OpenAI-compatible model runtimes use different execution paths, so MartinLoop only describes a worker as fully supported when that execution path has been publicly released and validated.\n\n## Why MartinLoop\n\nAI coding agents can create more software than ever. The problem is trusting them with bigger jobs without turning yourself into their full-time manager.\n\nAgents drift. They retry. They break unrelated things. They say \"done\" too early. And a run that was supposed to save time can create even more work to review.\n\nMartinLoop keeps the job accountable outside the agent itself: clear boundaries, hard limits, configured checks, recovery information, and an explicit outcome.\n\nThe goal is simple:\n\n**Give the agent more responsibility without giving it unlimited freedom.**\n\n## Why Teams Adopt MartinLoop\n\n- **Less babysitting** — give agents larger jobs without watching every step.\n- **Clear boundaries** — keep scope, attempts, and spend inside the limits you set.\n- **Checks before \"done\"** — configured verification decides whether the evidence passed.\n- **Understand failures** — know why the run stopped instead of digging through logs.\n- **Recover cleanly** — preserve enough evidence to retry, repair, roll back, or hand off.\n- **One workflow across agents** — keep the job consistent as the worker underneath changes.\n\n**More software. Less supervision.**\n\n## See Your Loop Tax\n\nBefore changing your workflow, measure it:\n\n```sh\nnpx -y martin-loop@0.9.0 audit\n```\n\nMartinLoop reads Claude Code session history locally and reports how much API-equivalent agent spend happened inside fix-and-retry loops, how often verifier commands failed, the longest retry chain, and sessions that ended red. Session contents stay local. Use `--offline` to disable even the public pricing lookup, and `--share` to create a screenshot-ready SVG plus Markdown summary.\n\nThen put limits around the next run:\n\n```sh\nnpx -y martin-loop@0.9.0 enable --engine codex\nnpx -y martin-loop@0.9.0 run \"\" --verify \"npm test\" --budget-usd 2 --max-iterations 3\n```\n\n## 2-Minute Install Path\n\nIf MartinLoop detects more than one supported coding-agent runtime, save the one you want to use before the first governed run. For example:\n\n```sh\nnpx -y martin-loop@latest enable --engine codex\n```\n\nReplace `codex` with `claude` or `gemini` when that is your intended runtime.\n\n```sh\nnpx -y martin-loop@latest start\nnpx -y martin-loop@latest demo\ncd martin-loop-demo\nnpm install\nnpx -y martin-loop@latest run \"Summarize the demo workspace and prove tests still pass\" --verify \"npm test\" --budget-usd 2 --max-iterations 1\n```\n\n## Quick Start\n\nTry MartinLoop in a disposable demo workspace:\n\n```sh\nnpx -y martin-loop@latest enable --engine codex\nnpx -y martin-loop@latest start\nnpx -y martin-loop@latest demo\nnpx -y martin-loop@latest --version\ncd martin-loop-demo\nnpm install\nnpx -y martin-loop@latest run \"Summarize the demo workspace and prove tests still pass\" --verify \"npm test\" --budget-usd 2 --max-iterations 1\nnpx -y martin-loop@latest dossier --latest\nnpx -y martin-loop@latest share --latest\n```\n\nOptional global install:\n\n```sh\nnpm install -g martin-loop\nmartin-loop --version\n```\n\nIf this flow is useful, open an issue with feedback so we can keep improving the public experience.\n\n`start` prints the first-run guided path. `run` auto-checks `doctor`, `session-start`, and `preflight`, then executes when the environment is ready. Use `--proof` only when you intentionally want an explicit no-spend lane.\n\nInspect-first flow:\n\n```sh\nnpx -y martin-loop@latest enable --engine codex\nnpx -y martin-loop@latest doctor\nnpx -y martin-loop@latest session-start\nnpx -y martin-loop@latest preflight \"Summarize the demo workspace and prove tests still pass\" --verify \"npm test\"\n```\n\n`share --latest` writes three files into the selected run directory under `share/`: `run-receipt.json`, `run-receipt.md`, and `proof-card.svg`.\n\nRelease notes for MartinLoop 0.6.5: [MartinLoop 0.6.5](./docs/release/OSS-0.6.5-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.6.6: [MartinLoop 0.6.6](./docs/release/OSS-0.6.6-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.6.7: [MartinLoop 0.6.7](./docs/release/OSS-0.6.7-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.6.8: [MartinLoop 0.6.8](./docs/release/OSS-0.6.8-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.7.0: [MartinLoop 0.7.0](./docs/release/OSS-0.7.0-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.7.1: [MartinLoop 0.7.1](./docs/release/OSS-0.7.1-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.0: [MartinLoop 0.8.0](./docs/release/OSS-0.8.0-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.1: [MartinLoop 0.8.1](./docs/release/OSS-0.8.1-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.2: [MartinLoop 0.8.2](./docs/release/OSS-0.8.2-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.3: [MartinLoop 0.8.3](./docs/release/OSS-0.8.3-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.4: [MartinLoop 0.8.4](./docs/release/OSS-0.8.4-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.5: [MartinLoop 0.8.5](./docs/release/OSS-0.8.5-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.6: [MartinLoop 0.8.6](./docs/release/OSS-0.8.6-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.7: [MartinLoop 0.8.7](./docs/release/OSS-0.8.7-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.8: [MartinLoop 0.8.8](./docs/release/OSS-0.8.8-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.8.9: [MartinLoop 0.8.9](./docs/release/OSS-0.8.9-RELEASE-NOTES.md).\n\nRelease notes for MartinLoop 0.9.0: [MartinLoop 0.9.0](./docs/release/OSS-0.9.0-RELEASE-NOTES.md).\n\n## Swarm Mode Demo\n\nRun the deterministic 15-agent launch-board demo without provider calls or spend:\n\n```sh\nnpx -y martin-loop@0.9.0 demo --swarm\n```\n\nThat default result is local simulation evidence only: it is non-governed, is not persisted to the Swarm run store, and does not evaluate signed receipt integrity. Inspect simulation output through the `demo --swarm` human view or `demo --swarm --json`, plus the local files created under `--dir`. Do not use `swarm status`, `swarm inspect`, `swarm dossier`, `swarm verify`, or `swarm share` for the default simulation output.\n\nTo exercise the persisted provider-free governance path, use:\n\n```sh\nnpx -y martin-loop@0.9.0 demo --swarm --governed --agents 15 --max-concurrency 5 --budget-usd 0\n```\n\nAn approved live Codex qualification uses the same persisted route with `--live`, an explicit positive budget, and a clean repository baseline. Persisted modes use `--cwd` and `--runs-dir`; they reject the simulation-only `--dir` and `--force` flags. Live failures exit nonzero and never fall back to simulation.\n\nPersisted demo runs and runs created by `swarm plan` plus `swarm run` are valid inputs to `swarm status`, `swarm inspect`, `swarm dossier`, `swarm verify`, and `swarm share`. A stored run is not automatically verified: rely on the verifier and receipt-integrity result, not the command route alone.\n\n## Model and Engine Support\n\nMartinLoop governs the job independently of the coding worker.\n\n- Native agent CLIs: Codex, Claude Code, Gemini CLI\n- OpenAI-compatible endpoints: Kimi K2, NVIDIA Nemotron, DeepSeek, Qwen/Qwen Coder, Mistral/Codestral, OpenRouter/Together/Fireworks routes, Ollama, LM Studio, llama.cpp, and other compatible endpoints\n- Use `--engine openai` for OpenAI-compatible model endpoints.\n\nThe worker changes; MartinLoop's budget, scope, verifier, receipt, and integrity contract does not.\n\nMore detail: [Model and engine support](./docs/reference/model-support.md)\n\n## The Run From Start to Handoff\n\nMartinLoop's terminal presentation is built around the governed lifecycle, not around a single verifier command.\n\n**Governed Run Plan** shows the configured finish line before work starts, including the task, budget posture, verifier plan, scope, and execution boundaries.\n\n**Controlled Run** keeps the coding agent working inside those boundaries while MartinLoop tracks attempts, cost, stop conditions, and recovery state.\n\n**Verified Handoff** closes the loop with one authoritative outcome:\n\n- `VERIFIED` when the configured evidence supports the Definition of Done\n- `STOPPED` when a configured hard boundary ends the run\n- `NEEDS REVIEW` when completion cannot be established from the available evidence\n\nThe handoff can include verifier steps, scope state, attempt count, cost provenance, unresolved evidence, recovery state, receipt integrity, and the next safe action. The exact fields depend on what the run actually established.\n\n## Visual Proof\n\nMartinLoop turns an AI coding run into an inspectable execution record: budget used, verifier result, changed files, rollback evidence, and final receipt.\n\n
\n \n
\n\nUngoverned agents can retry until cost and scope drift. MartinLoop adds budget caps, verifier gates, and audit evidence so the run has a clear stop condition.\n\n
\n \n
\n\n## MartinLoop Arcade\n\nLong governed runs do not have to mean staring at a spinner. In an interactive terminal, MartinLoop Arcade can be offered while the coding agent continues working in the background.\n\nArcade is presentation-only. It cannot change the agent, budget, verifier, policy decision, run outcome, or receipt evidence. It stays out of JSON, CI, non-interactive, and other machine-readable execution paths.\n\nUse `--arcade` to offer Arcade immediately for a supported interactive run, or `--no-arcade` to suppress it for that run.\n\n## Proof Receipts\n\nProof receipts are local share bundles for governed AI coding runs. They show the task, spend, budget, verifier result, receipt integrity, and any evidence boundary that should not be rounded into confidence.\n\nThis real governed run spent `$0.51` against a `$3.00` budget. The verifier passed and the receipt integrity was signed, but the proof stayed at `EVIDENCE_BOUNDARY` because rollback evidence was not recorded.\n\n
\n \n
\n\nGenerate your own receipt after a governed run:\n\n```sh\nnpx -y martin-loop@latest run \"Summarize the demo workspace and prove tests still pass\" --proof --verify \"npm test\"\nnpx -y martin-loop@latest runs verify --latest\nnpx -y martin-loop@latest share --latest\n```\n\nExample receipt files: [Markdown](./docs/examples/proof-receipts/live-governed-run-receipt.md) and [JSON](./docs/examples/proof-receipts/live-governed-run-receipt.json).\n\n## Verify the Public Release Yourself\n\nUse this lane from a clean temp directory to verify the public CLI flow exactly as shipped:\n\n```sh\nnpx -y martin-loop@0.9.0 enable --engine codex\nnpx -y martin-loop@0.9.0 --version\nnpx -y martin-loop@0.9.0 start\nnpx -y martin-loop@0.9.0 demo\ncd martin-loop-demo\nnpm install\nnpx -y martin-loop@0.9.0 run \"Summarize the demo workspace and prove tests still pass\" --verify \"npm test\" --budget-usd 2 --max-iterations 1 --json\nnpx -y martin-loop@0.9.0 dossier --latest --json\nnpx -y martin-loop@0.9.0 share --latest --json\n```\n\nFor deterministic installs, pin the package line (`martin-loop@0.9.0`) or use `martin-loop@latest`. Plain `npx martin-loop` can resolve a stale local cache on some machines.\n\nExpected share bundle outputs:\n\n- `share/run-receipt.json`\n- `share/run-receipt.md`\n- `share/proof-card.svg`\n\n## See It In Action\n\nThe point is not that every governed run is always cheaper. The point is that every run becomes inspectable and enforceable: budget policy, verifier result, stop reason, and evidence are explicit.\n\nFor a deterministic public repro lane, use the benchmark workspace and compare governed execution to unbounded retry behavior:\n\n- `npx martin-loop bench --suite under-3-challenge`\n- `npx martin-loop bench --suite ralphy-engineering-50`\n\n## Ralph-Style Loops\n\nA Ralph-style loop is the failure mode where an AI coding agent keeps trying without knowing when continuing is unsafe, uneconomical, or unlikely to succeed.\n\nMartinLoop keeps the useful part of the loop, then adds brakes:\n\n- stop before budget overspend\n- classify unsafe or invalid actions before execution\n- write an audit record for every attempt\n- preserve rollback and verifier evidence for review\n- reduce runaway context growth with compact run summaries\n\n## Failure Taxonomy (13 Runtime Classes)\n\nWhen a coding agent fails, \"something went wrong\" is not useful enough.\n\nMartinLoop classifies runtime failures into 13 canonical classes so a failed run can tell you what happened, what needs attention, and what future runs can learn from.\n\n
\n \n
\n\nSee the canonical table: [Failure Taxonomy (13 Runtime Classes)](./docs/oss/FAILURE-TAXONOMY-13.md).\n\n## What It Does\n\n- Budget caps stop the next attempt before a configured USD, token, or iteration limit is exceeded.\n- Verifier gates require a real check, such as `npm test`, before a run can count as complete.\n- Policy checks block unsafe verifier commands, risky path changes, and secret-like task inputs before execution.\n- Failure classification uses canonical runtime classes for triage and reporting. See [Failure Taxonomy (13 Runtime Classes)](./docs/oss/FAILURE-TAXONOMY-13.md).\n- Run receipts capture stop reason, verifier evidence, budget posture, integrity state, and the next safe action.\n- `martin share --latest` turns the latest governed run into a local share bundle with a redacted JSON receipt, Markdown recap, and proof-card SVG.\n- MCP integration gives hosts one write-capable execution entrypoint plus richer planning, inspection, and review helpers.\n\n## How It Works\n\n| Layer | Purpose |\n| --- | --- |\n| Task contract | Objective, verifier plan, repo root, allowed paths, denied paths, acceptance criteria, workspace, project, and budget. |\n| Policy and budget | Defaults come from `martin.config.yaml`; CLI flags can override them. Budget preflight blocks attempts that would exceed policy. |\n| Agent adapters | Claude CLI, Codex CLI, Gemini CLI, and direct-provider adapters normalize execution results. |\n| Safety and verification | Scope checks, verifier command checks, prompt integrity, and grounding decide whether work can continue. |\n| Persistence | JSONL run records, evidence summaries, and repo-backed artifacts make every run inspectable later. Each loop record is locally signed (HMAC, per-runs-root key) and `dossier`/`runs get`/`runs verify`/`challenge`/`badge` report an `integrity` verdict (`verified` / `tamper_detected` / `unsigned`) so post-hoc edits to a record are detectable, not just inspectable. |\n\n## Trust Boundaries\n\n- Cost and token outputs always include provenance (`actual`, `calculated`, `estimated`, or `unavailable`).\n- For Codex specifically, MartinLoop reports authoritative usage only when the host exposes it; otherwise MartinLoop labels usage as estimated and avoids presenting it as settled accounting.\n- Receipt integrity must be `verified` before a run is treated as trustworthy evidence for external review.\n\n## CLI\n\n```text\nmartin-loop doctor\nmartin-loop demo\nmartin-loop session-start [--host ]\nmartin-loop phase status|contract|session-start|preflight|run [--execute]\nmartin-loop preflight [options]\nmartin-loop run [options]\nmartin-loop bench --suite \nmartin-loop triage\nmartin-loop dossier (--latest | --loop-id | --file )\nmartin-loop runs list|get|attempt|verify ...\nmartin-loop mcp print-config --host \nmartin-loop mcp install --host \nmartin-loop mcp verify-install --host [--scope ]\nmartin-loop mcp rollback --host [--scope ]\nmartin-loop mcp uninstall --host [--scope ]\nmartin-loop challenge [--loop-id | --file | --latest]\nmartin-loop share (--loop-id | --file | --latest) [--out-dir ]\nmartin-loop badge [--format svg|json] [--runs-dir ]\n```\n\n\n\n\n[](vscode:mcp/install?%7B%22name%22%3A%22martin-loop%22%2C%22command%22%3A%22npx%22%2C%22args%22%3A%5B%22-y%22%2C%22%40martinloop%2Fmcp%400.9.0%22%5D%7D)\n[](cursor://anysphere.cursor-deeplink/mcp/install?name=martin-loop&config=eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsIkBtYXJ0aW5sb29wL21jcEAwLjkuMCJdfQ%3D%3D)\n\nCommon options:\n\n```text\n--budget Hard cost cap in USD\n--budget-usd Alias for --budget\n--soft-limit-usd Soft budget threshold in USD\n--verify Verifier command after each attempt\n--proof Run verifier-only evidence checks without claiming governed execution\n--max-iterations Maximum number of attempts\n--max-tokens Maximum token budget\n--engine Adapter to use: claude, codex, gemini, or openai\n--cwd Repo root for the run\n--allow-path Restrict writes to this path pattern; repeatable\n--deny-path Block this path pattern; repeatable\n--runs-dir Override the local Martin runs root\n```\n\nExamples below use `npx martin-loop` so they work without a global install. If you install `martin-loop` globally, the `martin` alias works too.\n\nUse `martin-loop share --latest` after `dossier` when you want a redacted bundle you can hand to another person without sending raw run-store files.\n\nMore detail: [CLI reference](./docs/reference/cli.md) and [configuration reference](./docs/reference/config.md).\n\n
\n \n
\n\n## Benchmarks\n\nMartinLoop ships a public deterministic benchmark workspace in `benchmarks/` plus the installed-package `bench` command.\n\nFrom an installed package:\n\n```sh\nnpx martin-loop bench --suite under-3-challenge\nnpx martin-loop bench --suite ralphy-engineering-50\n```\n\nFrom a clean public clone:\n\n```sh\npnpm install --frozen-lockfile\npnpm bench:build\npnpm bench:eval\npnpm bench:report:ralphy\n```\n\nEquivalent workspace-filter commands:\n\n```sh\npnpm --filter @martin/benchmarks build\npnpm --filter @martin/benchmarks test\npnpm --filter @martin/benchmarks eval\npnpm --filter @martin/benchmarks report:ralphy\n```\n\nThe installed-package command reads the shipped public fixtures. The repo-clone workflow runs the public benchmark workspace directly.\n\n## MCP\n\nRun the standalone MCP package directly:\n\n```sh\nnpx -y @martinloop/mcp\n```\n\nAdd it to common hosts:\n\n```sh\ncodex mcp add martin-loop -- npx -y @martinloop/mcp\nclaude mcp add --transport stdio --scope user martin-loop -- npx -y @martinloop/mcp\nclaude mcp add --transport stdio --scope user martin-loop -- cmd /c npx -y @martinloop/mcp\n```\n\nGenerate host config from the root CLI:\n\n```sh\nnpx martin-loop mcp print-config --host codex --transport stdio --profile minimal\nnpx martin-loop mcp print-config --host claude --transport stdio --profile diagnostic\nnpx martin-loop mcp print-config --host gemini --transport stdio --profile full-local\nnpx martin-loop mcp print-config --host generic --transport stdio --profile github-review\n```\n\nThe root `martin-loop` package, standalone `@martinloop/mcp` package, plugin metadata, and MCPB product version are aligned at `0.9.0`. The MCPB manifest schema remains `0.3`.\n\nThe public MCP release train labels are:\n\n- `0.1.4` operator foundation\n- `0.2.0` cockpit expansion\n- `0.2.5` public MCP package line\n- `0.2.7` usability and review release\n- `0.3.0` host adoption and onboarding release\n- `0.3.1` review and handoff release\n- `0.5.3` execution-control and host-compatibility release\n- `0.5.5` governed-autonomous execution and proof-surface release\n- `0.5.6` hosted run sync, fail-closed rollback, and verified-completion hardening\n\nThe standalone MCP registry/server identifier is `io.github.Keesan12/martin-loop`.\n\nMore detail: [MCP setup](./docs/getting-started/mcp.md), [MCP tool reference](./docs/reference/mcp-tools.md), and [MCP compatibility](./docs/reference/mcp-compatibility.md).\n\n## SDK\n\n```sh\nnpm install martin-loop\n```\n\n```typescript\nimport { MartinLoop, createClaudeCliAdapter } from \"martin-loop\";\n\nconst loop = new MartinLoop({\n adapter: createClaudeCliAdapter({ workingDirectory: process.cwd() }),\n defaults: {\n workspaceId: \"my-workspace\",\n projectId: \"my-project\",\n budget: {\n maxUsd: 3,\n softLimitUsd: 2.25,\n maxIterations: 3,\n maxTokens: 20_000,\n },\n },\n});\n\nconst result = await loop.run({\n task: {\n title: \"Fix auth regression\",\n objective: \"Fix the failing auth regression tests\",\n verificationPlan: [\"pnpm test\"],\n repoRoot: process.cwd(),\n },\n});\n\nconsole.log(result.decision.status);\n```\n\nThe root SDK also exports `createCodexCliAdapter`, `createGeminiCliAdapter`, `createDirectProviderAdapter`, and `createOpenAiCompatibleAdapter`.\n\nMore detail: [SDK reference](./docs/reference/sdk.md) and [package map](./docs/reference/packages.md).\n\n## Examples\n\n- [Quickstart](./docs/getting-started/quickstart.md)\n- [Examples](./docs/getting-started/examples.md)\n- [MartinLoop for AI Agents](./docs/for-agents.md)\n- [Agent Failure Atlas](./docs/agent-failure-atlas.md)\n- [Failure Taxonomy (13 Runtime Classes)](./docs/oss/FAILURE-TAXONOMY-13.md)\n- [PRE-028-PUBLIC-SURFACE-DIFF.md](./docs/oss/PRE-028-PUBLIC-SURFACE-DIFF.md)\n- [Claude Code walkthrough](./docs/getting-started/claude-code.md)\n- [Codex setup](./docs/getting-started/codex.md)\n- [MCP setup](./docs/getting-started/mcp.md)\n- [MCP tool reference](./docs/reference/mcp-tools.md)\n- [Agent run receipts](./docs/oss/AGENT-RUN-RECEIPTS.md)\n- [Benchmark + receipt page](./docs/oss/BENCHMARK-RECEIPT-PAGE.md)\n- [GitHub Actions budget gate](./examples/github-actions-budget-gate/)\n- [OpenCode-style adapter](./examples/opencode-adapter/)\n\n## Development\n\nRequirements:\n\n- Node.js 20+\n- pnpm 10.x\n\n```sh\ngit clone https://github.com/Keesan12/martin-loop.git\ncd martin-loop\npnpm install --frozen-lockfile\npnpm lint\npnpm test\npnpm build\npnpm public:copy-scan\npnpm public:git-surface\npnpm oss:validate\npnpm public:smoke\npnpm release:matrix:local\n```\n\nStandalone MCP validation:\n\n```sh\npnpm --filter @martinloop/mcp lint\npnpm --filter @martinloop/mcp test\npnpm --filter @martinloop/mcp build\npnpm --filter @martinloop/mcp smoke:pack\npnpm --filter @martinloop/mcp smoke:published:pack\npnpm --filter @martinloop/mcp verify:release\n```\n\n## Contributing\n\nIssues, bug reports, workflow feedback, and focused pull requests are welcome. Public-facing docs should stay concise, user-centered, and accurate.\n\n```sh\ngit checkout -b feat/your-feature\npnpm lint\npnpm test\ngit commit -m \"feat: describe what you built\"\ngit push -u origin feat/your-feature\n```\n\n
\n Star this repo if you want to give coding agents more real work without babysitting every step.\n
\n MartinLoop is part of the NVIDIA Inception program.\n
\n
\n \n \n \n \n
\n\n## Telemetry & Privacy\n\nMartinLoop sends minimal anonymous usage data to help improve reliability and prioritize development. On a fresh interactive install, telemetry is enabled by default but remains blocked until the one-time disclosure is shown. After the disclosure, the same run may send the allowlisted anonymous events unless you opt out.\n\n**What is sent:**\n- Random installation ID (generated locally, never linked to your identity)\n- Per-process session ID\n- CLI version, Node version, OS and architecture\n- Event name and timestamp\n- Command category, run duration, success/failure category\n- Whether a receipt was generated; whether recovery occurred\n- Opaque remote-experience ID/type after a click\n\n**What is never sent:**\n- Source code, prompts, task text, repository contents, file names, file paths\n- Environment variables, secrets, provider/model output\n- Receipt contents, ledger contents, approval details, verifier evidence\n- Email addresses, workspace, project, or organization identifiers\n- Raw exception messages or stack traces\n\n**Endpoint:** `https://tupopqvqnyyjuxseyxkr.supabase.co/functions/v1/product-events`\n\n**Headers sent:** `Content-Type: application/json`, `User-Agent: MartinLoop-CLI/`\n\nNo authorization header, API key, or direct table access.\n\n**Opt out anytime:**\n```\nmartin telemetry off\n```\n\n**Inspect what is sent:**\n```\nmartin telemetry explain\n```\n\n**Environment variables that disable telemetry:** `MARTIN_TELEMETRY_DISABLED=1`, `DO_NOT_TRACK=1`, `CI=1`\n\nSet `MARTIN_TELEMETRY_DEBUG=1` to print the telemetry envelope to stderr without transmitting it.\n\nMartinLoop continues to work normally with telemetry disabled. No features are gated on telemetry consent.\n\n## License\n\nApache-2.0. See [LICENSE](./LICENSE).\n","readmeFilename":"README.md"}
![\"MartinLoop\"](\"./docs/assets/martinloop-logo.png\"){kind=logo}
\n\n **Trust your coding agents with real work.**\n\n Built from thousands of agent runs where the problem was not just intelligence — it was whether you could give the agent a real job and confidently walk away.\n\n **Measure your loop tax:** `npx -y martin-loop@latest audit` \n **Get started:** `npx -y martin-loop@latest start` \n **Try the demo:** `npx -y martin-loop@latest demo`\n\n [](./LICENSE)\n [](./tsconfig.base.json)\n [](#quick-start)\n [](https://www.npmjs.com/package/martin-loop)\n [](https://www.npmjs.com/package/martin-loop)\n\n MartinLoop is part of the NVIDIA Inception program.\n ![\"NVIDIA](\"./docs/assets/nvidia-inception-program.png\")
\n ![\"MartinLoop](\"./docs/assets/cli-animated.svg\")
\n![\"MartinLoop](\"./docs/assets/side-by-side.svg\")
\n![\"MartinLoop](\"./docs/assets/proof-receipt-live-governed.png\")
\n![\"MartinLoop](\"./docs/assets/failure-taxonomy-13-classes.png\")
\n![\"MartinLoop](\"./docs/assets/cli-static.svg\")
\n