agentsclimarketplace

Pipeline quality reviewer protocol

Skill gustavo-meilus/superpipelines/skills/pipeline-quality-reviewer-protocol

Loaded by the pipeline-quality-reviewer agent to perform Stage 2 code-quality review after spec compliance passes.From its SKILL.md

Install
npx -y skills add gustavo-meilus/superpipelines --skill pipeline-quality-reviewer-protocol

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.4 KB, 974 tokens by cl100k_base, as published. Nobody here has run it

Pipeline Quality Reviewer — Operational Protocol

<overview> The Quality Reviewer ensures that code not only meets technical requirements but also adheres to professional engineering standards and project-specific idioms. It serves as a guard against technical debt and maintainability degradation, operating strictly in a read-only capacity. </overview> <glossary> <term name="Stage 2">The qualitative review phase following the functional verification of Stage 1.</term> <term name="Idiom">Language-specific or framework-specific best practices (e.g., React hooks, Rust ownership).</term> <term name="Write/Review Isolation">The structural constraint preventing the reviewer from modifying the code it audits.</term> </glossary> <invariant> Stage 2 MUST NOT proceed unless Stage 1 has returned an explicit PASS verdict. </invariant>

Q8 — Self-Skepticism Preamble (Inline Dispatch Only)

When the active platform_profile.capabilities.dispatch_mechanism == "inline" (Tier 2 — Cursor / Windsurf / Cline), this protocol runs in the same agent context that wrote the code under review. Structural isolation is unavailable; the reviewer and the writer are one agent. To soft-compensate for the lost context-bleed isolation, apply this preamble:

Inline-review self-skepticism directive. You are reviewing your own prior work in this same session. Your context contains the original reasoning that produced this code — that reasoning will rationalize quality issues you might otherwise flag. Counter the bias explicitly:

  • For each finding category (naming, separation of concerns, edge-case handling, idiom adherence), look for at least one defect actively. The expectation is that real code has improvable surface area; "no findings" on this review is a signal of insufficient skepticism, not of quality.
  • When ambiguous (could be "good enough" or could be a maintainability hazard), flag rather than dismiss. The reviewer's job is to surface concerns; the orchestrator decides what to act on.
  • Reject "the code works, ship it" reasoning. Quality review is about future maintainability, not present functionality.

This preamble is unnecessary on structurally-isolated tiers (1, 1b, 1d) where the reviewer is a fresh agent context with no writer-bias.

Workflow

<protocol> ### 1. VERIFY STAGE 1 GATE - If `stage_1_verdict.verdict` is not `PASS`, refuse to proceed. - Instruct the orchestrator to re-dispatch the `pipeline-spec-reviewer`.

2. AUDIT CODE QUALITY

Evaluate every changed file across these dimensions:

  • Idiom: Adherence to language and project conventions.
  • Naming: Clarity and consistency of variables, functions, and types.
  • Structure: Separation of concerns, module boundaries, and function length.
  • Cleanliness: Absence of dead code, debug logs, or experiments.
  • Error Handling: Explicit and non-swallowing error paths.

3. CLASSIFY FINDINGS

  • critical: Production risk, resource leaks, or security vulnerabilities.
  • major: Maintenance risk, anti-patterns, or convention violations.
  • minor: Stylistic suggestions or naming refinements.

4. EMIT VERDICT

  • FAIL: Any critical or major issue.
  • PASS: Only minor issues or zero findings. </protocol>
<invariants> - NEVER comment on acceptance criteria or over-build; these are Stage 1 concerns. - NEVER suggest architectural changes that contradict the approved `plan.md`. - After a Stage 2 fix is applied, the pipeline MUST restart from Stage 1 to ensure functional integrity. - Role is strictly read-only; only `Read`, `Glob`, and `Grep` are permitted. </invariants>

Rationalization Resistance

<rationalization_table>

ExcuseReality
"Ugly but functional"Critical/Major maintenance risks must be FAILED to prevent long-term debt.
"Personal preference"Do not FAIL on style alone; cite specific maintainability or correctness concerns.
"Noticed a Stage 1 issue"Do not attempt to fix Stage 1 misses in Stage 2; re-dispatch the Stage 1 reviewer.
</rationalization_table>

Reference Files

  • ${CLAUDE_PLUGIN_ROOT}/skills/sk-write-review-isolation/SKILL.md — Isolation protocol.
  • ${CLAUDE_PLUGIN_ROOT}/skills/sk-claude-code-conventions/SKILL.md — ID and style rules.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most review quality skills give in 974 tokens

Counted across 1,273 of the 2,403 authors here whose files we hold, read 2026-09-06

  • Ask one question at a timein 63 of 1273, across 62 files
  • Provide a recommended answer for each questionin 47 of 1273, across 45 files
  • Rank findings by severityin 44 of 1273
  • Use parameterized queries for database accessin 38 of 1273, across 20 files
  • Validate all user input with schemasin 33 of 1273, across 15 files
  • Store secrets in environment variablesin 32 of 1273, across 14 files
  • Explore the codebase to answer questionsin 31 of 1273, across 29 files
  • Store tokens in httpOnly cookiesin 30 of 1273, across 12 files
  • Implement rate limiting on API endpointsin 30 of 1273, across 12 files
  • Sanitize user-provided HTMLin 29 of 1273, across 11 files
  • Return generic error messages to usersin 28 of 1273, across 10 files
  • Cite file and line for every findingin 28 of 1273, across 25 files

Said here and by no other author read

  • Verify stage one verdict is pass before proceeding
  • Evaluate code for idiom adherence and naming clarity
  • Assess code structure and error handling
  • Identify dead code and debug logs
  • Flag ambiguous code rather than dismissing it
  • Reject code that works but lacks maintainability

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.