← Back to Discover

Mantyl

A local-first command-line tool that reads a repository and its Claude Code, Cursor or Codex history and produces a re-checkable project passport for handing over AI-built software.

Command-line toolsTesting & QA
Visit website ↗

What it does

Mantyl is a command-line tool for handing over software that was built with coding agents. It combines a repository with the history of agent sessions from Claude Code, Cursor and OpenAI Codex CLI to produce a project passport: a document describing the software's contents, how to run it, the decisions made while building it, which claims were demonstrated by running code, and which questions are still open. The aim is to replace the builder's assurances with a record the recipient can check. Every statement carries one of eight truth statuses with evidence references, from independently re-executed through agent-reported and inferred to contradicted. An LLM may interpret material, but the page says it can never mark anything verified. The workflow runs init, scan, verify and generate; verification executes install, build and test steps in an isolated Docker sandbox, and checks are recorded as skipped when Docker is unavailable. The output is a passport.json plus HTML and Markdown reports. The recipient runs a receive command that recomputes the passport digest, fingerprints the delivered files, compares the commit and re-executes the recorded checks in their own sandbox. Everything runs locally, nothing uploads without an explicit publish command, and transcript text passes through secret redaction first. It installs through npm, needs Node 20.12 or newer, and the repository is MIT licensed.

Capabilities

  • Builds a passport from a repository plus Claude Code, Cursor and Codex session history
  • Eight truth statuses so interpretation is never presented as proof
  • Sandbox verification of install, build and test steps through Docker
  • Receive command for recipients to re-verify a delivered passport
  • Secret redaction of transcript text and an open passport schema

When to use it

  • You are handing an AI-built project to a client or another team and want evidence, not assurances
  • You have inherited AI-built software and want to check what was claimed against what runs
  • You want to capture decisions made in coding-agent sessions before they are lost