Quality Scoring

A deterministic rubric for evaluating issue completion. The score drives closure decisions: auto-close, propose, or block.

Star Grading Scale

Scores are displayed as stars by default. Use --verbose to see numeric scores alongside stars.

Grade	Range	Label	Action
★★★★★	90-100	Exemplary	Auto-close eligible
★★★★	80-89	Strong	Auto-close eligible
★★★	70-79	Acceptable	Propose closure
★★	60-69	Needs Work	Propose closure
★	<60	Inadequate	Block closure

Display format: Configurable via .ccc-preferences.yaml:

scoring:
  display_format: stars    # stars (default) | numeric | letter | percentage

Format	Example Output
stars	★★★★ Strong (default)
numeric	85
letter	B+
percentage	85%

When --verbose is used, both the display format and the underlying numeric score are shown: ★★★★ Strong (85/100).

Pre-Step: Gather Issue Context Bundle

Before executing this skill, gather the issue context bundle (see issue-lifecycle/references/issue-context-bundle.md). Include comment thread health in the scoring assessment: are there unresolved questions in comments? Are dispatch results present but unreviewed? Comment thread state feeds into the Review dimension.

Scoring Dimensions

Dimension	Weight	What It Measures
Test	40%	Test coverage, tests passing, edge cases addressed
Coverage	30%	Acceptance criteria addressed, scope completeness
Review	30%	Review comments resolved, adversarial findings addressed, comment thread health

Dimension Rubrics

Test (40%)

Score	Criteria
100	All acceptance criteria have corresponding tests; all tests pass; edge cases covered
80	Core acceptance criteria tested; all tests pass; some edge cases missing
60	Tests exist but incomplete; all tests pass
40	Tests exist but some fail
20	Minimal tests, partial failures
0	No tests or all tests fail

Adjustments:

• -10 if test coverage decreased from baseline
• +10 if regression tests added for previously-discovered bugs
• For exec:quick mode: test requirement relaxed to 60 threshold (tests encouraged but not mandatory)

Coverage (30%)

Score	Criteria
100	Every acceptance criterion explicitly addressed with evidence
80	All criteria addressed; minor gaps in evidence
60	Most criteria addressed; 1-2 criteria partially met
40	Half of criteria addressed
20	Fewer than half addressed
0	No criteria addressed

Assessment method: Read the spec's acceptance criteria checklist. For each criterion, verify:

1. Implementation exists (code change, config change, or documented decision)
2. Evidence exists (test, screenshot, manual verification, or reviewer confirmation)

Review (30%)

Score	Criteria
100	All review comments resolved; all adversarial findings addressed or carry-forwarded
80	All blocking comments resolved; minor comments acknowledged
60	Blocking comments resolved; some minor comments unaddressed
40	Some blocking comments unresolved
20	Most blocking comments unresolved
0	No review conducted or all comments unresolved

Comment thread health adjustments:

• -10 if unresolved questions exist in the comment thread (questions posted but never answered)
• -5 if dispatch results are present in comments but have not been reviewed or acknowledged

Note: If no adversarial review was conducted (e.g., exec:quick mode), this dimension scores based on self-review or defaults to 70 (neutral).

Score Calculation

total = (test_score * 0.40) + (coverage_score * 0.30) + (review_score * 0.30)

Threshold Actions

Grade	Score	Action	Rationale
★★★★-★★★★★	80-100	Auto-close eligible	Quality gates met. Apply closure rules from issue-lifecycle.
★★-★★★	60-79	Propose closure	Close enough, but evidence gaps exist. List gaps in proposal.
★	0-59	Block closure	Specific deficiencies listed. Issue stays In Progress.

Important: Score >= 80 makes the issue eligible for auto-close, but the closure rules matrix from issue-lifecycle still applies. A ★★★★★ score on a human-assigned issue still requires human confirmation.

Scoring Output Format

Default (star grading)

## Quality Score — [Issue ID]

| Dimension | Grade | Evidence |
|-----------|-------|----------|
| Test (40%) | ★★★★ | 8/10 criteria tested, all passing, 2 edge cases missing |
| Coverage (30%) | ★★★★★ | 9/10 acceptance criteria met with evidence |
| Review (30%) | ★★★★★ | All 3 adversarial findings resolved |

**Overall: ★★★★★ Exemplary** — Auto-close eligible

### Gaps
- [ ] Edge case: empty input validation (test dimension)
- [ ] AC #7: partial — needs screenshot evidence (coverage dimension)

Verbose mode (--verbose)

## Quality Score — [Issue ID]

| Dimension | Grade | Score | Evidence |
|-----------|-------|-------|----------|
| Test (40%) | ★★★★ | 85 | 8/10 criteria tested, all passing, 2 edge cases missing |
| Coverage (30%) | ★★★★★ | 90 | 9/10 acceptance criteria met with evidence |
| Review (30%) | ★★★★★ | 100 | All 3 adversarial findings resolved |

**Overall: ★★★★★ Exemplary (90/100)** — Auto-close eligible

### Gaps
- [ ] Edge case: empty input validation (test dimension)
- [ ] AC #7: partial — needs screenshot evidence (coverage dimension)

Star Grading Conversion Rules

When converting a numeric score to stars:

1. Calculate the weighted total as before (0-100 scale)
2. Map to star grade using the threshold table above
3. Per-dimension scores also get star grades for display
4. The label (Exemplary/Strong/Acceptable/Needs Work/Inadequate) always accompanies stars

Letter grade mapping (when scoring.display_format: letter):

Grade	Range
A+	95-100
A	90-94
B+	85-89
B	80-84
C+	75-79
C	70-74
D	60-69
F	<60

Per-Project Customization

Projects can override default weights and thresholds in their CLAUDE.md:

# Quality scoring overrides
sdd_quality:
  weights:
    test: 0.50      # Higher test weight for safety-critical projects
    coverage: 0.30
    review: 0.20
  thresholds:
    auto_close: 90   # Stricter threshold
    propose: 70
  adjustments:
    exec_quick_test_floor: 50  # Lower test floor for quick mode

If no overrides exist, the defaults from this skill apply.

Integration with Closure

The /ccc:close command calls quality scoring as its first step:

1. Calculate score across all three dimensions
2. Convert to star grade (or configured display format)
3. Apply threshold rules to determine action (auto-close, propose, block)
4. If auto-close eligible, check closure rules matrix (issue-lifecycle skill)
5. Execute the appropriate action with the grade as evidence in the closing comment