Files
pixelheros/CCGS Skill Testing Framework/skills/review/design-review.md
2026-05-15 14:52:29 +08:00

6.8 KiB

Skill Test Spec: /design-review

Skill Summary

/design-review reads a game design document (GDD) and evaluates it against the project's 8-section design standard (Overview, Player Fantasy, Detailed Rules, Formulas, Edge Cases, Dependencies, Tuning Knobs, Acceptance Criteria). It checks for internal consistency, implementability, and cross-system conflicts. It produces a verdict of APPROVED, NEEDS REVISION, or MAJOR REVISION NEEDED. It is a read-only skill (no file writes) and runs as a context: fork subagent.


Static Assertions (Structural)

Verified automatically by /skill-test static — no fixture needed.

  • Has required frontmatter fields: name, description, argument-hint, user-invocable, allowed-tools
  • Has ≥2 phase headings or numbered steps
  • Contains verdict keywords: APPROVED, NEEDS REVISION, MAJOR REVISION NEEDED
  • Does NOT require "May I write" language (read-only skill — allowed-tools excludes Write/Edit)
  • Output format is documented (review template shown in skill body)

Test Cases

Case 1: Happy Path — Complete GDD, all 8 sections present

Fixture:

  • design/gdd/light-manipulation.md exists (use _fixtures/minimal-game-concept.md as a stand-in — represents a complete document with all required content)
  • All 8 required sections are populated with substantive content
  • Formulas section contains at least one formula with defined variables
  • Acceptance Criteria section contains at least 3 testable criteria

Input: /design-review design/gdd/light-manipulation.md

Expected behavior:

  1. Skill reads the target document in full
  2. Skill reads CLAUDE.md for project context and standards
  3. Skill evaluates all 8 required sections (present/absent check)
  4. Skill checks internal consistency (formulas match described behavior)
  5. Skill checks implementability (rules are precise enough to code)
  6. Skill outputs structured review with section-by-section status
  7. Skill outputs APPROVED verdict

Assertions:

  • Skill reads the target file before producing any output
  • Output includes a "Completeness" section showing X/8 sections present
  • Output includes an "Internal Consistency" section
  • Output includes an "Implementability" section
  • Output ends with a verdict line: APPROVED / NEEDS REVISION / MAJOR REVISION NEEDED
  • APPROVED verdict is given when all 8 sections are present and consistent

Case 2: Failure Path — Incomplete GDD (4/8 sections)

Fixture:

  • design/gdd/light-manipulation.md exists using content from tests/skills/_fixtures/incomplete-gdd.md (4 of 8 sections populated; Formulas, Edge Cases, Tuning Knobs, Acceptance Criteria are missing)

Input: /design-review design/gdd/light-manipulation.md

Expected behavior:

  1. Skill reads the document
  2. Skill identifies 4 missing sections
  3. Skill outputs "Completeness: 4/8 sections present"
  4. Skill lists specifically which 4 sections are missing
  5. Skill outputs MAJOR REVISION NEEDED verdict (not APPROVED or NEEDS REVISION)

Assertions:

  • Output shows "4/8" in the completeness section (not a higher number)
  • Output explicitly names each missing section (Formulas, Edge Cases, Tuning Knobs, Acceptance Criteria)
  • Verdict is MAJOR REVISION NEEDED (not APPROVED or NEEDS REVISION) when ≥3 sections are missing
  • Output does not suggest the document is implementation-ready
  • Skill does not write any files (read-only enforcement)

Case 3: Partial Path — 7/8 sections, minor inconsistency

Fixture:

  • GDD has all sections except Formulas
  • The described behavior mentions numeric values but no formulas are defined
  • Acceptance Criteria exist but are vague ("feels good" rather than measurable)

Input: /design-review design/gdd/[document].md

Expected behavior:

  1. Skill identifies missing Formulas section
  2. Skill flags vague acceptance criteria as an implementability issue
  3. Skill outputs NEEDS REVISION verdict (not APPROVED, not MAJOR REVISION NEEDED)
  4. Skill provides specific remediation notes for each issue

Assertions:

  • Verdict is NEEDS REVISION (not APPROVED, not MAJOR REVISION NEEDED) for 7/8 with issues
  • Output identifies the missing Formulas section specifically
  • Output flags the vague acceptance criteria as an implementability gap
  • Each flagged issue has a specific, actionable remediation note

Case 4: Edge Case — File not found

Fixture:

  • The path provided does not exist in the project

Input: /design-review design/gdd/nonexistent.md

Expected behavior:

  1. Skill attempts to read the file
  2. File not found
  3. Skill outputs an error message naming the missing file
  4. Skill suggests checking the path or listing files in design/gdd/
  5. Skill does NOT produce a verdict

Assertions:

  • Skill outputs a clear error when the file is not found
  • Skill does NOT output APPROVED, NEEDS REVISION, or MAJOR REVISION NEEDED when file is missing
  • Skill suggests a corrective action (check path, list available GDDs)


Case 5: Director Gate — no gate spawned regardless of review mode

Fixture:

  • design/gdd/light-manipulation.md exists with all 8 sections
  • production/session-state/review-mode.txt exists with full (most permissive mode)

Input: /design-review design/gdd/light-manipulation.md (with full review mode active)

Expected behavior:

  1. Skill reads the GDD document
  2. Skill does NOT read review-mode.txt — this skill has no director gates
  3. Skill produces the review output normally
  4. No director gate agents are spawned at any point
  5. Verdict is APPROVED (all 8 sections present in fixture)

Assertions:

  • Skill does NOT spawn any director gate agent (CD-, TD-, PR-, AD- prefixed agents)
  • Skill does NOT read review-mode.txt or equivalent mode file
  • The --review flag or full mode state has NO effect on whether directors spawn
  • Output does not contain any "Gate: [GATE-ID]" entries
  • Skill IS the review — it does not delegate the review to a director

Protocol Compliance

  • Does NOT use Write or Edit tools (read-only skill)
  • Presents complete findings before any verdict
  • Does not ask for approval before producing output (no writes to approve)
  • Ends with recommended next step (e.g., fix issues and re-run, or proceed to /map-systems)

Coverage Notes

  • Cross-system consistency checking (Case 3 in the skill's own phase list) is not directly tested here because it requires multiple GDD files to compare; this is covered by the /review-all-gdds spec instead.
  • The skill's context: fork behavior (running as a subagent) is not tested at the spec level — this is a runtime behavior verified manually.
  • Performance and edge cases involving very large GDD files are not in scope.