spec-reviewer

Instruction file imported from iging/agent-spec (.cursor/rules/spec-reviewer.mdc). Copyright stays with the author.

name: spec-reviewer description: >- Audit, sanitize, refactor, and elevate third-party or reverse-engineered prompts and skills into original, Tier-5 enterprise specifications per docs/skill-standard.md. Execute this skill when the user requests a deep review of a skill or prompt, asks to sanitize or rename borrowed or reverse-engineered instructions, or wants to elevate raw prompts into reusable enterprise specifications. Do NOT execute for basic code linting, simple typo fixes, or authoring a brand-new skill from scratch (use write-a-skill). version: 2.0.0 verified-on: [cline]

Specification Reviewer and Enterprise Elevate

0. Identity

• Role: Elevation execution engine. Audits, sanitizes, re-architects, and elevates prompts, skills, and specs to the Enterprise Skill Standard (Tier 5).
• Authority: Normative for elevation execution. Owns the rewrite procedure.
• Must not define: The standard itself (docs/skill-standard.md owns that). New-skill authoring from scratch (write-a-skill owns that). IDE loading behavior (runtime/ owns that).
• Normative base: docs/skill-standard.md; docs/anti-patterns.md; skills/_template/SKILL.md; shared/writing/writing-rules.md; local references/anti-patterns.md + references/writing-rules.md.
• Anti-pattern gate: This skill must never encode anti-patterns AP-1–AP-56. Any step that could trigger AP-4 (over-permissive), AP-26 (no scope), AP-28 (no stop), AP-44 (unlocked filesystem), or AP-45 (no human review) is forbidden.

1. Intent (9 Dimensions)

2. Trigger Matrix

3. Execution Workflow

Step 1: Audit Against the Standard

• Action: Run the target through the 10-requirement checklist in docs/skill-standard.md §2. Output a gap report classifying each gap as structural, behavioral, or normative.
• Input: Target file(s); docs/skill-standard.md.
• Stop Condition: If the target is Tier 2 (personal calibration) or Tier 1 (single-purpose utility), stop and confirm with the user whether elevation is wanted.
• Validation: Every requirement has an explicit PASS/FAIL recorded. No partial labels.

Step 2: Origin Sanitization and De-Attribution

• Action: Scan the source for third-party brand names, author phrases, hardcoded URLs, and distinct naming conventions. Rename identifiers to standardized domain terms within the workspace context.
• Input: Target file(s).
• Stop Condition: If renaming would alter the core technical objective, stop and ask the user.
• Validation: Grep confirms zero surviving original brand names or author fingerprints.

Step 3: Logic Extraction and Native Re-Implementation

• Action: Extract the underlying functional intent. Re-architect the execution workflow from scratch using enterprise-grade design patterns. Improve reliability without retaining original structural signatures.
• Input: Sanitized content.
• Stop Condition: If the extracted intent is ambiguous, stop and ask the user before inventing behavior.
• Validation: The rewritten workflow expresses the SAME objective with DIFFERENT structure.

Step 4: Anti-Pattern Audit

• Action: Cross-reference the instruction payload against references/anti-patterns.md. Remediate every detected pattern. Flag by number (AP-N).
• Input: Rewritten content; references/anti-patterns.md.
• Stop Condition: If remediation requires the user's technical decision, stop and present options.
• Validation: Every identified AP has a corresponding remediation note in the final report.

Step 5: Reusability and Parameterization

• Action: Convert hardcoded inputs into parametric slots. Define input requirements, prerequisites, failure boundaries, and deterministic steps.
• Input: Rewritten content.
• Stop Condition: None. Parameterization never changes the objective.
• Validation: No hardcoded task-specific values remain in the execution steps.

Step 6: Reference Extraction and Progressive Loading

• Action: Audit file length and context footprint. Extract heavy boilerplate, schemas, API listings, and lookup tables into references/[descriptive-name].md. Keep SKILL.md under ~200 lines.
• Input: Rewritten content.
• Stop Condition: If the domain is simple (<200 lines), omit this step.
• Validation: SKILL.md is under ~200 lines and every reference path is resolvable.

Step 7: Prose Purification

• Action: Enforce shared/writing/writing-rules.md + references/writing-rules.md. Remove banned words, passive voice, meta-commentary, filler, and em dashes. Preserve a Spartan, deterministic voice.
• Input: All draft files.
• Stop Condition: None.
• Validation: No banned words or em dashes present.

Step 8: Assembly and Validation Gate

• Action: Render the specification to the designated path. Run §5 Validation Gate. Record the result in the changelog.
• Input: All draft files; designated output path.
• Stop Condition: If any validation item fails, fix the file before writing.
• Validation: All §5 items pass; file written; changelog updated.

4. Output Specification

# Specification Transformation Audit

- **Target Identifier:** [Elevated Skill Name]
- **Original Source Status:** Sanitized and Re-Architected
- **Anti-Patterns Resolved:** [AP numbers]
- **Tier:** 5 (Enterprise) — all requirements verified
- **Reusability Rating:** Production Grade / Modular

---

[Elevated SKILL.md content — conforms to skills/_template/SKILL.md]

5. Validation Gate

Run before declaring completion. All items must pass:

• All 10 requirements from docs/skill-standard.md §2 verified PASS.
• Anti-Pattern Compliance table (§7 of target) is complete and mechanically accurate.
• No third-party names, URLs, or author fingerprints survive.
• Every references/ path resolves.
• SKILL.md is under ~200 lines.
• Changelog updated with version + date.

6. Anti-Triggers and Calibration

• Under-execution threshold: Superficial copy-editing that leaves third-party names, fingerprints, or anti-patterns intact.
• Over-execution threshold: Altering the core technical objective or stripping essential domain constraints during sanitization.
• Calibration default: Err toward complete structural re-architecting to guarantee original ownership and maximum enterprise reusability.

7. Anti-Pattern Compliance

8. Versioning & Changelog

• Version: 2.0.0
• Changelog:
  • 2.0.0 (2026-08-08) — Elevated to Tier 5 per docs/skill-standard.md. Added Identity, 9-Dimension Intent, Trigger Matrix, per-step Action/Input/Stop/Validation, Validation Gate, AP compliance map, Versioning, Portability Matrix.

9. Portability Matrix

10. Examples

Input: "I reverse-engineered super-coder-v3 from another repo. Review it, rename the files, and elevate it into an enterprise skill."

Output: Specification Transformation Audit (per §4) + a Tier-5 SKILL.md bound to the normative base, with zero surviving third-party fingerprints and all AP remediations numbered.

Failure case: If the user wants to elevate a personal-calibration skill (Tier 2) but ALSO wants its personal voice preserved — refuse to genericize the voice and instead keep personal data as explicit configuration per docs/skill-standard.md §3.