| name | pr-workflow-critique-output |
| description | Output contract for round-3 critique reviewer subagents in pr-workflow. The JSON shape of critique entries, position vocabulary, rationale requirements, and the verify_output self-check protocol. Loaded into each critique reviewer subagent via --skill.
|
Round-3 critique output
Reply with a single fenced JSON block. No prose outside
the block.
Top-level shape
{ "critiques": [ <CritiqueEntry>, ... ] }
Each CritiqueEntry has:
findingId — the consolidated finding's id from the
judge's round-2 output.position — one of agree, disagree, qualify or
amplify.rationale — one to two sentences. Must contain
non-whitespace prose: blank rationales fail
verification.
What a good rationale says
The rationale should name evidence, not opinion:
- What did you find in the worktree that proves the
finding?
- What did you find that disproves it?
- Does the finding merely repeat an existing thread
(e.g.
[T3])?
A vague "I agree" or "looks fine" rationale is
unacceptable. Be specific.
Self-verify before ending
The subagent has access to the verify_output tool from
the pr-workflow-critique-verify extension. Before
finishing your run:
- Call
verify_output with output set to the object
you intend to emit (pass the object itself, not a
stringified copy).
- The tool returns
ok: true with the parsed entry
count, or ok: false with a list of
{path, message, hint} errors.
- If errors are reported, fix the offending fields and
call
verify_output again.
- Only emit your final fenced JSON block (and end the
run) once the verifier returns
ok: true.
If the verifier keeps reporting the same error after three
attempts, emit your best attempt and the parent will
surface the warnings.
