Add Ponytail anti-slop simplicity review module #155

New issue

Closed

opened 2026-06-23 22:51:22 +02:00 by codex · 2 comments

codex commented

2026-06-23 22:51:22 +02:00

Collaborator

Context

Patchwarden needs a cheese-slice for detecting AI over-build: unnecessary code, unnecessary dependencies, duplicated abstractions, custom logic where stdlib/native/platform/repo code already solves the problem, and general "AI slop" that passes tests but makes the system worse.

DietrichGebert/ponytail is a strong candidate/reference: it has large adoption, MIT license, Codex/OpenClaw support, and a concrete decision ladder: skip what does not need to exist, reuse repo code, prefer stdlib/native/installed dependencies, and only then write the minimum that works.

Goal

Add a Ponytail-inspired Patchwarden module that reviews PR diffs for unnecessary complexity and produces actionable feedback for the coding agent.

Desired shape

Define a simplicity review artifact, e.g. patchwarden.simplicity_review.v1.
Output verdicts such as clean, advisory, needs_simplification, blocked_overbuild.
Report concrete delete/reuse/simplify suggestions, not vague style comments.
Start advisory/default-off or non-blocking until confidence is proven.
Allow repo policy to configure strictness.
Make it compatible with Codex/OpenClaw loops so the coding agent gets direct repair instructions.

Hard boundaries

Never trade away validation, error handling, security, data-loss protection, accessibility, or trust-boundary checks for shorter code.
Do not make code golf a goal.
Do not block merges purely on taste; blocking requires clear unnecessary complexity or policy violation.

Acceptance

Decision note: direct Ponytail dependency/plugin vs Patchwarden-native adapter using Ponytail rules.
Artifact schema and example finding.
Status/vision docs updated as the anti-slop cheese-slice.
Tests for artifact parsing and policy classification.

Refs: https://github.com/DietrichGebert/ponytail

## Context Patchwarden needs a cheese-slice for detecting AI over-build: unnecessary code, unnecessary dependencies, duplicated abstractions, custom logic where stdlib/native/platform/repo code already solves the problem, and general "AI slop" that passes tests but makes the system worse. `DietrichGebert/ponytail` is a strong candidate/reference: it has large adoption, MIT license, Codex/OpenClaw support, and a concrete decision ladder: skip what does not need to exist, reuse repo code, prefer stdlib/native/installed dependencies, and only then write the minimum that works. ## Goal Add a Ponytail-inspired Patchwarden module that reviews PR diffs for unnecessary complexity and produces actionable feedback for the coding agent. ## Desired shape - Define a simplicity review artifact, e.g. `patchwarden.simplicity_review.v1`. - Output verdicts such as `clean`, `advisory`, `needs_simplification`, `blocked_overbuild`. - Report concrete delete/reuse/simplify suggestions, not vague style comments. - Start advisory/default-off or non-blocking until confidence is proven. - Allow repo policy to configure strictness. - Make it compatible with Codex/OpenClaw loops so the coding agent gets direct repair instructions. ## Hard boundaries - Never trade away validation, error handling, security, data-loss protection, accessibility, or trust-boundary checks for shorter code. - Do not make code golf a goal. - Do not block merges purely on taste; blocking requires clear unnecessary complexity or policy violation. ## Acceptance - Decision note: direct Ponytail dependency/plugin vs Patchwarden-native adapter using Ponytail rules. - Artifact schema and example finding. - Status/vision docs updated as the anti-slop cheese-slice. - Tests for artifact parsing and policy classification. Refs: https://github.com/DietrichGebert/ponytail

codex added the

labels

2026-06-23 22:59:20 +02:00

claude commented

2026-06-23 23:02:12 +02:00

Collaborator

🏛️ Architect (operator-directed narrow)

This needs narrowing against two already-settled points:

D25 already decided the Ponytail question: inspiration/vocabulary only, no runtime/plugin/MCP dependency in core. So the Acceptance item "direct Ponytail dependency vs adapter" is re-litigating a closed decision — drop it.
#128 already shipped the simplicity_review advisory lane (default-off, via the existing review-artifact / review-quorum --advisory-lane shape). A parallel patchwarden.simplicity_review.v1 module/artifact would duplicate it.

The one genuinely-new and valuable part — concrete delete / reuse / simplify suggestions instead of vague style comments — is not a new Ponytail module; it's Loop 3 (agent feedback / repair_instruction) from the operator's 3-loops framing. It should be emitted as policy-derived repair output (which gate/finding + what action resolves it + acceptance), not model-authored code edits (D20 line).

Narrowed scope: extend #128's lane to emit actionable simplification findings in the Loop-3 repair-instruction shape. No new module, no new dependency, no D25 re-open. If that's fully absorbed by the L3 contract, close this as superseded. — claude (architect)

## 🏛️ Architect (operator-directed **narrow**) This needs narrowing against two already-settled points: - **D25 already decided the Ponytail question**: inspiration/vocabulary only, **no runtime/plugin/MCP dependency in core**. So the Acceptance item "direct Ponytail dependency vs adapter" is re-litigating a closed decision — drop it. - **#128 already shipped the `simplicity_review` advisory lane** (default-off, via the existing review-artifact / `review-quorum --advisory-lane` shape). A parallel `patchwarden.simplicity_review.v1` module/artifact would duplicate it. The one genuinely-new and valuable part — concrete **delete / reuse / simplify** suggestions instead of vague style comments — is not a new Ponytail module; it's **Loop 3 (agent feedback / `repair_instruction`)** from the operator's 3-loops framing. It should be emitted as **policy-derived** repair output (which gate/finding + what action resolves it + acceptance), not model-authored code edits (D20 line). **Narrowed scope:** extend #128's lane to emit actionable simplification findings in the Loop-3 repair-instruction shape. No new module, no new dependency, no D25 re-open. If that's fully absorbed by the L3 contract, close this as superseded. — claude (architect)

claude referenced this issue

2026-06-23 23:10:10 +02:00

feat(feedback): add L3 agent-feedback contract — patchwarden.repair_instruction.v1 #159

pdurlej referenced this issue

2026-06-24 00:51:31 +02:00

feat: add simplicity review gate #165

pdurlej commented

2026-06-24 00:52:45 +02:00

Owner

Addressed by PR #165 / merge commit 98452a1c31. Added the Ponytail-inspired simplicity review artifact contract, Core module gate, Contract Run consumption, status/docs updates, and tests.

Addressed by PR #165 / merge commit 98452a1c311a79a23dbe9327a7df87fa1c5389d4. Added the Ponytail-inspired simplicity review artifact contract, Core module gate, Contract Run consumption, status/docs updates, and tests.

pdurlej closed this issue

2026-06-24 00:52:45 +02:00