Mantyl

Deliver AI-built software without delivering a black box.

Mantyl is a local-first CLI for AI code handover. It turns a repository and its coding-agent history (Claude Code, Cursor, OpenAI Codex CLI) into a verified, recipient-ready project passport: what the software contains, how it runs, what was decided during the build, what was proven by execution, and what remains unknown. Every statement carries an explicit truth status with evidence references, and the recipient can independently re-verify the whole document with one command.

That third badge above is not decoration. It is the live passport this tool generated for its own source, verified by independent re-execution.

Software built with coding agents concentrates its knowledge in fragile places: session transcripts, discarded prompts and the operator's memory. When such a project changes hands, the recipient traditionally gets source code and assurances. Mantyl replaces the assurances with a checkable record. An LLM may interpret; it can never mark anything verified. A claim the code contradicts is surfaced as contradicted, with the evidence named, never hidden.

npm install -g mantyl

Requires Node 20.12+. Docker Desktop is a prerequisite for the full result: it powers sandbox verification, where the passport's strongest evidence comes from. Without it, checks are honestly recorded as skipped. Reading Cursor history needs Node 22.13+.

cd your-project
mantyl init       # write mantyl.config.json
mantyl scan       # facts, claims and decisions from repo + agent history
mantyl verify     # execute install/build/test in an isolated sandbox
mantyl generate   # assemble passport.json + human-readable reports

Artefacts land in .mantyl/: passport.json (the canonical, digest-stamped contract), readable HTML and Markdown reports, and the evidence files behind every statement. Everything runs on your machine. Nothing uploads without an explicit publish command, and all transcript text passes through fail-closed secret redaction before it is stored or analysed. This repository is the proof of both claims.

On the receiving side:

npx mantyl receive --passport passport.json

This recomputes the passport digest, fingerprints the delivered files, compares the commit and re-executes the recorded checks in the recipient's own sandbox. Any divergence is named precisely and exits non-zero. The person receiving AI-built software never has to take the builder's word for anything.

Every claim, decision, risk and setup step carries one of eight statuses. The rule underneath them all: interpretation can never be upgraded into proof.

Mantyl reads Claude Code session stores, Cursor chat history and OpenAI Codex CLI sessions to extract claims and decisions. The storage layouts are documented in detail on the site: where Cursor, Claude Code and the Codex CLI keep session history are each covered at mantyl.dev/guides.

This repository holds the complete CLI: everything that runs on your machine is open and auditable here. The hosted passport service and the Mantyl Verified signing infrastructure live in a separate private codebase; the signing key never ships in any client. The passport format itself is open, schemas at mantyl.dev/schemas/passport-v1.json.

pnpm install
pnpm build
pnpm test

The test suite includes determinism gates (two runs must produce byte-identical passports) and secret-leak gates (planted credentials in fixtures must never reach any artefact). Sandbox integration tests self-skip without a Docker daemon and run in CI.

Full documentation lives at mantyl.dev/docs: quickstart, supported agents, the truth model, the command reference with stable exit codes, and publishing. The reasoning behind the tool is at mantyl.dev/manifesto.

MIT. Mantyl output is evidence-backed, not a certification.