Correctness findings carry a test the author runs, and every decline cites its evidence #488

Parent

#486

What to build

The review shows its evidence, and the author answers with evidence, as #486 sets out:

• The skill asks for evidence. The thirdshift-code-review skill's two sub-agent briefs ask for a test or a command, written out in full, that fails if a finding is right. This applies to every finding that says behaviour is wrong, whichever axis raises it. Spec findings keep quoting the requirement, and Standards findings keep citing the standard or naming the smell. The briefs' word limits grow enough to hold the evidence.
• The author runs it. The shared review instruction tells the author to run each such test or command as written before deciding the finding. A failure obliges the fix, and the test is kept. A pass allows a decline that cites the run. A passing run counts only when it exercises that finding; a green suite doesn't.
• Declines cite evidence. Each entry in the pull request's Unaddressed findings carries what refutes the finding, in place of "at least a one-line reason": a line of code, a plan decision, an ADR, or the run.
• A call that isn't the agent's becomes an issue. When the author declines a finding because a Spec or the Day shift must decide, it files a needs-triage issue, unless an open issue already covers it, and links the issue from the entry. The issue says what the finding is, its evidence, which pull request raised it, and why the call isn't the agent's. It carries no other label, so it pauses nothing.
• The glossary matches. The Unaddressed finding entry in CONTEXT.md says each entry carries what refutes the finding, or the issue filed for a decision the agent may not make.

Acceptance criteria

• The skill's briefs ask for a written-out test or command for each finding that says behaviour is wrong, on either axis, and leave Spec quoting and Standards citing as they are.
• The briefs' word limits leave room for the evidence.
• The prompts page shows each of the four prompts telling the author to run the evidence as written. A failure means a fix and a kept test; a pass allows a decline that cites the run.
• The prompts page shows each Unaddressed finding carrying what refutes it, and "one-line reason" is gone.
• The prompts page shows a decision the agent may not make becoming a needs-triage issue, unless an open issue covers it, linked from its entry. The issue says what the finding is, its evidence, the pull request that raised it, and why the call isn't the agent's.
• The issue carries no label but needs-triage.
• The wording says a passing run counts only when it exercises that finding.
• The Unaddressed finding entry in CONTEXT.md matches.

Blocked by

• One review instruction for the prompts that run the review #487