agent-engineering-toolkit 1.18.0

Proof-carrying workflows for coding agents: bind tests to exact code and detect stale evidence.

Agent Engineering Toolkit

Your coding agent says tests passed. AET proves which code was actually tested.

AET provides proof-carrying workflows for coding agents. It binds test runs, changed files, artifacts, and agent claims into portable evidence, then tells you when that proof no longer applies.

Try the stale-proof demo

After v1.18.0 is published:

uvx --from agent-engineering-toolkit aet demo stale-proof

Expected result:

1. Test command executed                    PASS
2. Proof matches the tested source          EXACT_MATCH
3. Source changed without rerunning tests   RELEVANT_FILES_CHANGED

Demo result: PASS

The test really passed. Then a relevant source file changed without rerunning it, so AET correctly stopped applying the old proof to the current code.

The demo is local and deterministic after package download. It uses Git and Python's standard-library unittest; it makes zero network or LLM calls.

Install

uv tool install agent-engineering-toolkit
aet --version

The PyPI package currently remains at v1.11.1. Do not advertise the v1.18 demo command until PyPI or the exact GitHub Release wheel is verified.

Choose the smallest surface

Question	Command
Are Agent instructions usable?	aet quick check .
Does a diff fit the task?	aet quick scope . --base main --intent aet.intent.json
Did this command run on these files?	aet quick proof --output proof.json --relevant-path src/app.py -- <argv>
Does old proof still apply?	aet quick fresh --proof proof.json
What should change without editing?	aet plan context ...

AET is not another coding agent. It does not replace tests or CI, turn missing evidence into PASS, or auto-edit, commit, push, merge, or release.

• Source and documentation
• Five-minute start
• Status and authority
• Security

Python 3.11+ · MIT License · no product telemetry