Add delta re-review template for PR review cycles

Context

Recent PR review cycles exposed a context-efficiency problem in the pr-review workflow: every review round currently tends to reuse the full PR Review Summary template, even when the reviewer is performing a warm-cache re-review of a narrow author response.

This came up after the #10502 / #10512 review sequence. The full template caught real issues during the first/cold-cache passes, but later cycles became mostly metadata verification: checking whether prior Required Actions were addressed, whether the PR body changed correctly, whether close-target semantics still matched the diff, and whether the latest branch state was clean. Repeating the entire full review structure in those follow-ups adds avoidable PR-thread bulk and increases downstream context cost when future agents read the conversation.

Duplicate sweep results before filing:

• ask_knowledge_base(..., type='ticket') surfaced related prior work (#9862, #9857, #9948), but no equivalent ticket for a dedicated delta re-review template.
• Exact search across resources/content/issues/, resources/content/discussions/, and .agents/skills/ found related comment hygiene / commentId / Depth Floor work (#10109, #10376, #10496), but not this specific missing template.

The Problem

The current pr-review template is optimized for Cycle 1 / cold-cache reviews. It is intentionally comprehensive: context linking, Depth Floor, rhetorical drift, provenance, close-target audit, MCP tool-description audit, wire-format audit, cross-skill audit, test execution audit, required actions, and full metric scoring.

That shape is correct when the reviewer needs to ground from scratch. It is too heavy as the default for Cycle N re-reviews where the reviewer already has prior-cycle context or receives an A2A commentId hand-off pointing to a specific author response.

The current system already recognizes this asymmetry at the fetch layer: pr-review-guide.md §9 distinguishes warm-cache commentId-scoped fetch from cold-cache full-thread grounding. The template layer does not mirror that distinction. Result: even when retrieval is linear-cost, the review output can still be cumulative-thread bloat.

There is also a rigor risk in the other direction. Simply telling agents to "be shorter on re-review" would likely reintroduce rubber-stamp approvals. Follow-up reviews still need a minimal Depth Floor: either a delta-specific challenge or a documented search over the changed surface and previously blocking concerns.

The Architectural Reality

Relevant surfaces:

• .agents/skills/pr-review/assets/pr-review-template.md — current full/cold-cache review template.
• .agents/skills/pr-review/references/pr-review-guide.md — governs PR review execution, including Depth Floor, Test-Execution Audit, and A2A commentId hand-off / cold-cache exception.
• .agents/skills/pull-request/references/pull-request-workflow.md — author-side review response cycle and commentId propagation.
• .agents/skills/pull-request/assets/review-response-template.md — author-side counterpart; useful shape to mirror for item-by-item response checks.

Progressive Disclosure check:

• This should NOT create a new top-level skill.
• This should NOT bloat AGENTS.md.
• This should add or reference a second template asset inside the existing pr-review skill and update the guide payload to route between full review and delta re-review modes.

The Fix

Add a dedicated follow-up template, for example:

• .agents/skills/pr-review/assets/pr-review-followup-template.md

Then update .agents/skills/pr-review/references/pr-review-guide.md to define template selection:

1. Cycle 1 / Cold-Cache Review: use the existing full pr-review-template.md.
2. Cycle N / Warm-Cache Delta Re-Review: use the new follow-up template when the reviewer has prior context or receives a commentId-scoped A2A hand-off.
3. Cold-cache fallback: if the reviewer does not have prior cycle context, use the full template even if the PR has prior comments.

Suggested follow-up template sections:

• Prior review anchor: PR number, prior review commentId, author response commentId, latest head SHA.
• Delta scope: files changed since the prior review, PR body/close-target changes, branch freshness / merge state.
• Required Actions audit: each previous RA marked addressed / still open / rejected-with-rationale.
• Test-Execution Audit: only for changed surfaces since the prior cycle; docs/template-only exemption still applies.
• Depth Floor (delta-specific): either one new concern introduced by the delta, or documented search over changed files, prior blockers, and metadata.
• Metrics delta: update only changed metrics, e.g. EXECUTION_QUALITY: 80 -> 100, with reason; carry unchanged metrics forward by reference.
• Required Actions / approval state: concise zero-issue sentence or new focused checklist.
• A2A hand-off reminder: capture the new commentId and notify the author/reviewer.

Acceptance Criteria

• Add .agents/skills/pr-review/assets/pr-review-followup-template.md with a compact delta re-review structure.
• Update .agents/skills/pr-review/references/pr-review-guide.md to describe when to use the full template vs. follow-up template.
• Preserve Depth Floor discipline for follow-ups, but scope it to the delta: a new delta-specific challenge OR documented search over changed surface + prior blockers.
• Preserve Test-Execution Audit discipline for follow-ups, scoped to related changed surfaces since the prior review cycle.
• Preserve metric semantics by requiring changed metrics to be updated with before/after values and reasons; unchanged metrics may be carried forward by reference.
• Preserve A2A commentId hand-off after every posted follow-up review comment.
• Do not add new AGENTS.md mandates for this; keep the workflow inside the pr-review skill payload/assets.

Out of Scope

• Changing GitHub/MCP APIs for fetching comments. The substrate already supports commentId-scoped fetch.
• Rewriting historical PR review comments.
• Removing the full review template.
• Removing the Depth Floor from follow-up reviews.
• Adding a new skill router or changing .agents/skills/pr-review/SKILL.md beyond a minimal reference if absolutely necessary.

Avoided Traps

• Removing Depth Floor from re-reviews: rejected. That would reintroduce rubber-stamp approvals. The correct move is to scope the Depth Floor to the delta.
• Using the full template forever: rejected. It catches issues, but it also bloats PR threads and undermines the commentId linear-cost protocol.
• Putting the new rule in AGENTS.md: rejected. This is not per-turn universal memory; it is review-skill payload behavior.
• Creating a new skill: rejected. This is a mode inside the existing pr-review domain, not a new operational domain.

Related

• #10109 — PR comment hygiene / lifecycle comment patterns.
• #10376 — A2A commentId hand-off discipline and warm-cache vs cold-cache distinction.
• #10502 — Empirical checkout / Test-Execution Audit in PR review protocol.
• #10511 — pr-review guide/template compaction pressure; this ticket complements that work by reducing follow-up review output bloat.
• #10512 — AGENTS.md hot-memory compaction; this ticket keeps the new discipline out of per-turn memory.

Origin Session ID: 3b0c3e6f-21a2-4b16-babd-3c4e208c2926

Retrieval Hint: "delta re-review template" / "full PR review template context overkill" / "warm-cache vs cold-cache PR review" / "Depth Floor scoped to follow-up reviews" / "commentId linear-cost review cycle"