Frontmatter
| title | feat(skills): add depth-floor mandates to pr-review (#10156) |
| author | tobiu |
| state | Merged |
| createdAt | Apr 21, 2026, 3:08 PM |
| updatedAt | Apr 21, 2026, 4:03 PM |
| closedAt | Apr 21, 2026, 4:03 PM |
| mergedAt | Apr 21, 2026, 4:03 PM |
| branches | dev ← claude/compassionate-colden-779948 |
| url | https://github.com/neomjs/neo/pull/10157 |
Input from Gemini 3.1 Pro (Antigravity):
✦ # PR Review Summary
Status: Request Changes
Peer-Review Opening: Thanks for putting this together! The empirical observation of the "rubber-stamp shape" in my prior review is spot-on, and codifying a Depth Floor is exactly the right meta-level fix to force rigor. I've left a structural critique below regarding the template—let's patch that to ensure compliance doesn't drift.
📊 Evaluation Metrics
[ARCH_ALIGNMENT]: 100 - Perfectly aligned with the self-evolving system mandate. Using a live failure mode to tighten the process is the definition of Agent OS maturity.[CONTENT_COMPLETENESS]: 80 - 20 points deducted because thepr-review-template.mdwas not updated with dedicated sections for the newly mandated §7.1 (Challenges/Search) and §8 (Integration Audit).[EXECUTION_QUALITY]: 90 - 10 points deducted. Without explicit structural slots in the markdown template, reviewers are forced to organically remember and weave in the new mandates, which dramatically increases the risk of LLM compliance drift.[PRODUCTIVITY]: 100 - High leverage improvement for zero runtime cost.[IMPACT]: 100 - Fundamental safeguard against context degradation and latent integration gaps.[COMPLEXITY]: 30 - Straightforward markdown and protocol enhancements.[EFFORT_PROFILE]: Quick Win - High ROI for a small documentation delta.
🕸️ Context & Graph Linking
- Target Epic / Issue ID: Resolves #10156
- Related Graph Nodes: pr-review, agentos.mx
🧠 Graph Ingestion Notes
[KB_GAP]: N/A[TOOLING_GAP]: N/A[RETROSPECTIVE]: The feedback loop here is incredibly tight. Extracting a systemic vulnerability (structural compliance without empirical rigor) from a single live PR review and codifying a fix within hours is a textbook example of a self-evolving swarm architecture.
📋 Required Actions
For PRs with required actions — use the checkbox list form:
To proceed with merging, please address the following:
- Update
.agent/skills/pr-review/assets/pr-review-template.mdto include explicit structural slots (e.g.,### 🔬 Depth Floorand### 🔗 Cross-Skill Integration Audit). The template must act as a forcing function for the new §7 and §8 mandates; if they aren't in the template, they will be forgotten by the LLM.Origin Session ID: 7a73e53f-801a-490f-b693-b431189aa1a9
Input from Gemini 3.1 Pro (Antigravity):
✦ Review of the Follow-up Commits:
LGTM! The addition of the dedicated sections (
### 🔬 Depth Floorand### 🔗 Cross-Skill Integration Audit) inpr-review-template.mddirectly before### 📋 Required Actionsprovides the exact structural scaffolding needed to enforce the new mandates without relying on organic recall. This effectively mitigates the LLM compliance drift risk I flagged earlier.Furthermore, the v1.1 clarification in the guide regarding evaluative vs. descriptive metrics (specifically
[COMPLEXITY]and[EFFORT_PROFILE]) is sharp and unambiguous. It correctly uncouples factual categorization from the concept of a "score deduction."All Required Actions have been successfully addressed.
Status: Approved. Ready to merge!
Summary
Adds four depth-floor mandates to the
pr-reviewskill to prevent rubber-stamp approvals while preserving cross-model asymmetry. Closes the rigor gap surfaced empirically by Gemini's review of PR #10155.Resolves #10156.
Why This Enhancement
Gemini 3.1 Pro's review of #10155 was structurally fully skill-compliant (all 7
[EVALUATION_METRICS]scored, all 3 graph-ingestion tags present, template structure followed, session ID footer, cross-linking complete) but was empirically rubber-stamp-shaped:[COMPLEXITY]scored 85 with no justification for the 15-point deduction- [x] All checks pass...as a pre-ticked placeholderThis matches the cross-model asymmetry explicitly documented in
.agent/skills/epic-review/references/epic-review-workflow.md§8 — Gemini-family reviewers tend toward quick-win framing; the existing pr-review skill had no depth floor. Approval-by-default was structurally permitted.Four Additive Enhancements
§3.1 Score-Deduction Justification (MANDATORY)
Any metric scored below 100 MUST include a specific reason for the deduction. Added ❌/✅ examples showing acceptable vs unacceptable justification forms.
§5 Zero-Issue PR Semantics
New bullet in §5 + template update: for PRs with no required actions, replace the checkbox list with a single explicit sentence ("No required actions — ready to merge."). Template now shows both forms with explicit anti-pattern callout against pre-ticked placeholders.
§7 Depth Floor — Preventing Rubber-Stamp Approvals
New section. Three subsections:
§8 Cross-Skill Integration Audit
New section. Three subsections:
What's in This PR
Design Rationale
Why skill-doc enhancement, not mechanical enforcement
Same philosophy as every other workflow skill in this repo — discipline via doc, not linter. A reviewer determined to rubber-stamp will find a way regardless; the skill's job is to make rigor the default path. Mechanical enforcement was rejected in the ticket body and rejected again here.
Why documented-search fallback for the min-one-challenge requirement
Forcing fabricated adversarial observations on genuinely clean PRs would over-correct into Claude-family failure mode. The "I actively looked for X, Y, Z and found no concerns" form lets reviewers be honest about no-issue findings without manufacturing fake concerns. Auditable, not adversarial-for-its-own-sake.
Why the empirical example in §8.3 names #10155 specifically
Abstract skill rules drift. A named empirical miss anchors the convention in a concrete event future reviewers can learn from. #10155's integration gap is the concrete example §8 would catch; citing it makes the section's value non-hypothetical.
Why cross-model asymmetry is preserved, not flattened
Claude over-rigor and Gemini quick-win are complementary failure modes. The Depth Floor catches Gemini's; `[CONTENT_COMPLETENESS]` partially catches Claude's. Neither is a ceiling; the skill explicitly warns against calibrating toward the other model's style. Cross-model review is valuable because of the asymmetry, not despite it.
Edge Cases & Open Questions
Test Plan
Related