Verify Behavior

---
name: verify-behavior
description: Execute acceptance tests for one spec and record only functional failures.
metadata:
  aiddbot-kind: primitive
user-invocable: true
---
# verify-behavior

Your goal is to evaluate one spec against its acceptance criteria and record any failures.

Run `node .agents/aidd/aidd.mjs run acceptance [--project {project}]` from the spec branch. Never edit code, tests, or contracts, and never run `lint`, `unit`, `quality`, or a test tool yourself. A requirement without an acceptance test tagged or titled with its global ID is a failure like a failing test.

A failure that an open item of `node .agents/aidd/aidd.mjs debt list`, recorded before this spec, already describes is pre-existing: when it explains every failure, record green with `--preexisting <D IDs>`. Every other failure counts, including shipped behavior the spec does not replace, and adding debt never excuses one. When only a product decision can tell whether a failure is expected, return that question without recording the evaluation, so no revision is spent.

For `amber` or `red`, write `{Product_Folder}/specs/{spec_key}/verification.md` from `assets/verification.template.md` with only failing or blocked acceptance checks and their evidence, marking each failing test tagged with another spec as a regression. Then run `node .agents/aidd/aidd.mjs eval verification <status> "<summary>"` from the spec branch; it records the next revision at the current commit, drops the report when green, and commits the record. Never edit `control.json`, `debt.json`, or anything under `.aiddbot/` by hand: when a command refuses, fix what its error names and run it again.

The result is a recorded green evaluation or a finding-only acceptance report.

The core commits as `docs(verification): record acceptance`.