Add shared prompt quality gate for plateaued agent-review workflows

[Copilot]

Agent quality/effectiveness has been stuck around 61/62 for weeks, which points to prompt design issues rather than more engine/runtime bug-fixing. This change adds a reusable quality rubric and applies it to the analyzer and the recurring low-signal reviewer/gate workflows that most directly influence prompt quality feedback loops.

• Shared quality gate

  • Added shared/quality-gate.md as a reusable prompt fragment.
  • Forces a silent self-check before emitting output:
    • completeness
    • accuracy
    • actionability
    • freshness of tool/workflow guidance
  • Instructs agents to prefer noop over low-confidence or generic output.

• Agent Performance Analyzer

  • Added an explicit Prompt Improvement Initiative section.
  • Shifts the analyzer from generic quality reporting to prompt-specific diagnosis:
    • audit the bottom 10 agents by quality score
    • classify deficiencies as generic framing, missing self-assessment, stale guidance, or low actionability
    • cite prompt/output/run evidence for each deficiency
  • Makes Matt Pocock Skills Reviewer, Impeccable Skills Reviewer, and Design Decision Gate explicit redesign-vs-deprecate candidates when they continue to underperform.
  • Requires a prioritized prompt-improvement backlog in the report output.

• Reviewer prompt tightening

  • Updated Matt Pocock Skills Reviewer and Impeccable Skills Reviewer with concrete success criteria.
  • Emphasizes:
    • high-signal findings only
    • changed-line specificity
    • concrete fixes
    • noop instead of filler praise
  • Removed ambiguous guidance in the Matt Pocock reviewer around optional follow-up outputs by making approval behavior explicit.

• Design Decision Gate

  • Added a gate-specific quality bar so decisions are easier to act on.
  • Requires the gate to clearly classify outcomes as:
    • exempt
    • ADR-backed
    • divergent
    • insufficient context
  • Tightens expectation that the gate cites evidence, gives one next action, and stops after the single final safe output.

• Compiled workflow updates

  • Recompiled the affected workflow lockfiles so the prompt changes take effect in Actions.

Example of the shared gate now imported by affected workflows:

## Quality Gate

Before you emit any safe output or final review, run one silent self-check and revise once if needed.

Verify all of the following:

1. **Completeness**
2. **Accuracy**
3. **Actionability**
4. **Freshness**

---

Generated by 👨‍🍳 PR Sous Chef · 8.12 AIC · ⌖ 9.29 AIC · ⊞ 3.7K · ◷
_{Comment /souschef to run again}

[Copilot]

Pull request overview

Adds a shared “quality gate” prompt fragment and applies it across key agent-review workflows to improve prompt-driven quality feedback loops and reduce low-signal outputs.

Changes:

• Added a reusable shared/quality-gate.md rubric and wired it into the affected compiled workflows.
• Tightened reviewer/gate prompts with explicit success criteria and improved approval/check-run guidance.
• Updated the Agent Performance Analyzer prompt to drive a bottom-10, prompt-deficiency-focused improvement backlog.

Show a summary per file

File	Description
.github/workflows/shared/quality-gate.md	Introduces a reusable rubric for completeness/accuracy/actionability/freshness and noop guidance.
.github/workflows/mattpocock-skills-reviewer.md	Adds success criteria and imports the shared quality rubric.
.github/workflows/mattpocock-skills-reviewer.lock.yml	Recompiled workflow prompt to include the shared quality gate.
.github/workflows/impeccable-skills-reviewer.md	Adds success criteria and imports the shared quality rubric.
.github/workflows/impeccable-skills-reviewer.lock.yml	Recompiled workflow prompt to include the shared quality gate.
.github/workflows/design-decision-gate.md	Adds a gate-specific quality bar and imports the shared quality rubric.
.github/workflows/design-decision-gate.lock.yml	Recompiled workflow prompt to include the shared quality gate.
.github/workflows/agent-performance-analyzer.md	Adds a prompt-improvement initiative and imports the shared quality rubric.
.github/workflows/agent-performance-analyzer.lock.yml	Recompiled workflow prompt to include the shared quality gate.

Review details

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 9/9 changed files
• Comments generated: 5
• Review effort level: Low

[github-actions]

🤖 PR Triage — Run §28741077056

Field	Value
Category	feature
Risk	🟡 Medium
Priority Score	50/100 — impact:25 urgency:10 quality:15
Action	🚀 fast_track

Rationale: Adds shared quality gate prompt rubric targeting stalled 61/62 agent effectiveness score. Targets 9 workflow files. 1× COMMENTED only — needs human review of prompt design before merge.

Generated by 🔧 PR Triage Agent · 91.2 AIC · ⌖ 11.9 AIC · ⊞ 5.5K · ◷

[gh-aw-bot]

@copilot please run the pr-finisher skill, address unresolved review comments, and rerun checks once the branch is up to date.

Generated by 👨‍🍳 PR Sous Chef · 8.12 AIC · ⌖ 9.29 AIC · ⊞ 3.7K · ◷
_{Comment /souschef to run again}

[Copilot]

@copilot please run the pr-finisher skill, address unresolved review comments, and rerun checks once the branch is up to date.

...

Implemented and pushed fixes in 4d3c6b5 to address the unresolved review threads (removed redundant nested runtime-import usages and aligned the Design Decision Gate classification wording). I also reran local checks (make fmt, make lint, make test-unit, make test, make recompile) successfully. CI on this updated head will need a maintainer-triggered rerun.

[github-actions]

Hey @Copilot 👋 — nice work introducing the shared quality-gate.md fragment and threading it through the analyzer, both skills reviewers, and the design decision gate. The motivation is clearly articulated (quality score plateau at ~61/62) and the scope is well-contained.

One item worth addressing before merge:

• No test coverage — the diff touches 9 files (4 prompt sources + 4 compiled lockfiles + 1 new shared fragment) but includes no test changes. If there are workflow snapshot tests, integration test fixtures, or golden-file comparisons for these prompts, they should be updated to reflect the new quality-gate.md import and the revised prompt text. If the project has no automated tests for workflow prompt files yet, a brief note in the PR body acknowledging that gap (or a follow-up issue) would help reviewers understand the risk surface.

If you'd like a hand, you can assign this prompt to your coding agent:

Check whether the repository has any snapshot tests, golden files, or integration test fixtures for the following workflow prompts:
- .github/workflows/agent-performance-analyzer.md
- .github/workflows/design-decision-gate.md
- .github/workflows/impeccable-skills-reviewer.md
- .github/workflows/mattpocock-skills-reviewer.md
- .github/workflows/shared/quality-gate.md

If test fixtures exist, update them to reflect the new shared quality-gate import and the revised prompt content from PR #43527. If no tests exist, open a follow-up issue titled "Add prompt snapshot tests for shared/quality-gate.md and affected workflows" describing what should be covered.

Generated by ✅ Contribution Check · 237.6 AIC · ⌖ 35.8 AIC · ⊞ 6.2K · ◷

[github-actions]

🎉 This pull request is included in a new release.

Release: v0.82.3