pr-workflow-stack-review-output

name	pr-workflow-stack-review-output
description	Output contract for stack-wide reviewer subagents in pr-workflow. The JSON shape with per-PR and cross-PR findings, perPr key rules, cross-PR span fields, and the verify_output self-check protocol. Loaded into each stack-review reviewer subagent via --skill.

Stack-review output

Reply with a single fenced JSON block. No prose outside the block.

Top-level shape

{
  "perPr": {
    "<prNumber>": [ <Finding>, ... ],
    ...
  },
  "crossPr": [ <CrossFinding>, ... ]
}

perPr

• Keys are PR numbers encoded as strings (e.g. "101", "205"). They must match the pattern /^[1-9][0-9]*$/.
• Include every PR in the stack as a key, using an empty array [] when that PR has no findings.
• Each finding has the same core shape as round 1: location, label, subject, discussion, plus the optional decorations, severity, confidence, threadRelation fields.

crossPr

True cross-PR observations only. A finding belongs in crossPr when its reasoning spans multiple PRs in the stack and can't be expressed under any single PR.

Each cross-PR finding has:

• The core finding fields (location, label, subject, discussion, etc.).
• homePrNumber — the PR where this finding should post.
• spans — non-empty array of PR numbers (integers, e.g. [248, 250]) the finding refers to.

Discipline

Walk each PR in order. Before moving to the next PR, decide every finding that belongs to the current PR and place it under perPr["<number>"]. After the last PR, add only true cross-PR observations under crossPr.

Self-verify before ending

The subagent has access to the verify_output tool from the pr-workflow-stack-review-verify extension. Before finishing your run:

1. Call verify_output with output set to the object you intend to emit (pass the object itself, not a stringified copy).
2. The tool returns ok: true with the parsed item count, or ok: false with a list of {path, message, hint} errors.
3. If errors are reported, fix the offending fields and call verify_output again.
4. 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.