agent-engineering-toolkit 1.11.1

An evidence-driven Agent engineering quality and control layer for delivery proof, deterministic diagnosis, and bounded asset evolution.

Agent Engineering Toolkit

Stop coding Agents from claiming “tests passed” without verifiable proof.

AET is a local CLI that records exact command execution, declared artifacts, human intent and the Git workspace snapshot. It can then check whether that evidence still matches the current code before a handoff or release.

Agent claim → exact execution evidence → live freshness check → human decision

Install

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

Record one proof

Declare the expected proof in aet.intent.json, then run the exact command through Trace:

aet trace --proof unit-tests --intent aet.intent.json \
  --artifact reports/pytest.txt \
  --output .aet/evidence/trace.json \
  -- python -m pytest -q

aet evidence receipt --report .aet/evidence/trace.json

If the workspace changes after the test, the historical command remains a success while current freshness becomes FAIL. Missing evidence remains UNKNOWN; AET never markets it as a pass.

Use the smallest surface

Question	Command
Are Agent instructions and Skills structurally usable?	aet audit . --strict
Is a diff inside the approved change contract?	aet review . --base main --intent aet.intent.json
Did this exact command produce this artifact?	aet trace … -- <argv>
Does recorded proof still match the workspace?	aet evidence receipt --report <trace.json>

AET is opt-in and normally off. It does not replace your Agent, tests or CI, and it never auto-adopts, commits, pushes or releases a governance candidate.

• Source and full documentation
• 60-second stale-proof case study
• Stability contract
• Security and retention boundaries
• Contributing

Python 3.11+ · MIT License