verification-before-completion

name	verification-before-completion
description	Use when a session is about to say DONE, 完成, ready to push, fixed, or ✅ outside an active /execute execution loop, and the same response does not already include fresh verify output with exit code and last 5 lines.

verification-before-completion

Do not claim completion without fresh verification evidence. Because: outside /execute, there is no step packet or checkbox gate to prove the work was actually checked.

Skip when already inside active /execute per-step execution. /execute has its own step packet, scope check, RED/GREEN evidence, verify, commit, and checkbox loop.

Trigger

Use this before saying:

• DONE
• 完成
• ready to push
• fixed
• ✅
• "it works"

Only trigger when the same response does not already contain fresh verify output.

Required action

1. Identify the relevant verify command.
2. Run it.
3. Report:
   • command
   • exit code
   • last 5 lines
4. Only then claim completion.

Evidence convention

For plan execution, prefer /execute evidence files. For freeform work, chat evidence is acceptable if it includes the command, exit code, and last 5 lines.

Related

• ${BDC_HOME:-$HOME/.bdc}/skills/execute/references/tdd-cycle.md § 6
• ${BDC_HOME:-$HOME/.bdc}/skills/execute/SKILL.md

Verification

• The final response includes command, exit code, and last 5 lines.
• If no command can be identified, the response says verification could not be run instead of claiming done.

Red Flags

• The response says "should work" or "fixed" without command output.
• The only evidence is an earlier run from before the latest edit.
• The task changed behavior but no test, lint, typecheck, or script self-test ran.

Common Rationalizations

Rationalization	Response
"This was just docs/config, no verify needed."	Advisory unless the file is behavior-rule config. Incident: standards/debugging.md 2026-04-26 records silent skip branches in BDC automation; behavior-rule files need executable checks when possible.
"The previous run passed."	Do not claim done. Incident: standards/debugging.md 2026-04-01 says recursive automation looked alive while core actions were not actually running; fresh evidence must be current.