FIELD NOTE / X
Exit zero is not enough evidence.
The short film, the complete written thought, and the evidence behind it.
The X edition will be linked here after its public post is verified.
Exit zero is not enough evidence.
Day 06 · 2026-10-03 · X
Short video caption
A post-tool hook can faithfully record an exit code and still miss an unrun acceptance suite. Trusted CI should bind revision, contract version, nonzero relevant tests and an independent postcondition; a deliberate fault must turn red. Illustrative. #EricFieldNotes
Full written post / accessible read
Post-tool hooks are valuable for recording what happened. But if the agent ran a wrapper that swallowed a failing test's exit status, the hook can record a clean process exit while the relevant suite never passed.
In this illustrative case, a broad script runs one small unit test and skips the protected acceptance suite. The post-tool message says tests complete. A reviewer sees a green line, but no evidence that the intended contract was executed.
A trusted CI job records the exact commit and contract version, parses the test report for nonzero relevant cases, and checks the independent outcome. Introduce a deliberate fault and require that suite to fail. A natural-language log is not the verdict.
Keep the acceptance runner outside the agent's writable worktree. Reject wrong revisions, missing reports, zero relevant tests and a surviving deliberate fault. Do this because a post-tool hook observes a call; it cannot make weak evidence strong after the fact.
#EricFieldNotes
X thread draft
Attach the video to the first post. The subsequent text adds detail; the full read above is also the accessible transcript. Review the thread in the live composer before sending.
A post-tool hook can faithfully record an exit code and still miss an unrun acceptance suite. Trusted CI should bind revision, contract version, nonzero relevant tests and an independent postcondition; a deliberate fault must turn red. Illustrative. #EricFieldNotes
In this illustrative case, a broad script runs one small unit test and skips the protected acceptance suite. The post-tool message says tests complete. A reviewer sees a green line, but no evidence that the intended contract was executed.
A trusted CI job records the exact commit and contract version, parses the test report for nonzero relevant cases, and checks the independent outcome. Introduce a deliberate fault and require that suite to fail. A natural-language log is not the verdict.
Keep the acceptance runner outside the agent's writable worktree. Reject wrong revisions, missing reports, zero relevant tests and a surviving deliberate fault. Do this because a post-tool hook observes a call; it cannot make weak evidence strong after the fact.
Evidence and boundary
On-screen boundary: ILLUSTRATIVE CI RECEIPT. The sources below support documented mechanisms and specifications; illustrative scenarios are not presented as measured incidents.