Mtm Contract — AI skill for Claude Code
Stop your AI agent from writing the wrong code — a contract-first discipline, with an installable Claude Code plugin.
How to install Mtm Contract
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open jewanchen/mtm-contract and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Mtm Contract does
Stop your AI agent from writing the wrong code — a contract-first discipline, with an installable Claude Code plugin.
Alternatives in AI
- Memory Forge Rs — Stop resetting satisfying AI chats — edit the memory instead 474 ★
- Mancode — AI coding agent harness 358 ★
- Styleseed — Claude Code & Cursor design system 325 ★
README
MTM Contract
**MTM — *machine to machine*.** A contract is a handoff format. Today a person usually writes it for an agent; the same file works when one agent hands work to another. What has to be written down does not change either way.
**A framework for preventing hallucination and architectural drift in agentic AI coding.** A discipline for containing agentic AI's two recurring failure modes. Markdown-only, zero dependencies, Apache 2.0. From **Vast Intelligence Limited**.
**Project page: [nicemeetapp.com/mtm](https://nicemeetapp.com/mtm/)** — the same material as a page rather than a repository, in English and Traditional Chinese.
A day you have probably had
A join starts returning nothing for records that visibly exist. The obvious move is to reason about it: stale cache, wrong identity, the data layer inferring a column type wrong. Three hypotheses, three changes, three rebuilds, no information — every layer above the database was internally consistent, so no amount of reading application code was going to surface it. One query against the database's own catalogue of columns ended it: the two sides of the join were different types.
The contract does not make you smarter. It makes you write down what must be true and how you checked it, **before** the agent generates anything — so that query happens in minute one instead of minute forty.
That is the whole idea: **put the cheap checks before the expensive generation.**
Start here — paste this into your agent
This is the method at working scale. Put it in `CLAUDE.md`, `.cursorrules`, or your system prompt. Nothing to install, no lock-in.
## Contract-first workflow
Before writing implementation code, classify the task by trigger — not by
how risky it feels. Any hit promotes it; never argue a task back down.
- Typo, copy, styling, version bump, one small file → just do it.
- One module, no trigger below → write three fields first (see below).
- Crosses a module bound
Related Skills
Vibe Tuning
Stop fixing the same AI mistake twice. Methodology + Claude Code skill that turns wrong AI results into perman
Answer Contract
Four rules that stop your AI agent from padding, rambling, and shipping one mediocre answer: Implicit-Need Fil
Claude Code Discipline Skills
Seven Claude Code skills that stop an LLM agent asserting what it has not measured. Each born from a dated pro
Verify Before Blocking
A portable discipline that stops AI agents from turning unverified assumptions into stop-work instructions. Si
Agent Wormhole
Stop self-replicating prompt payloads from writing themselves into your AI agent's config files
Git AI
Stop writing commit messages by hand. AI-powered conventional commits from your staged changes — powered by Cl
Related Agents
Kit Session Auditor
Use proactively when a contract is implemented and its session is unaudited — the stop-gate names it with the
Part 11: Auto-Capture Hook (Stop Manually Writing Memory)
By Terp - Terp AI Labs --- The vault system from Part 9 is powerful, but there's a gap: you have to manually w
Aiskills Incident Investigator
Investigates live production incidents end to end — triages impact and start time, draws the blast radius, mit