Evidence Heavy Evaluator

Run a deterministic repo evaluation and emit auditable artifacts in test-output.

Workflow

1. Choose inputs:

• target_dir: repo or subdirectory to evaluate.
• profile: readiness, maintainability, or release-readiness.
• depth: quick or deep.
• execute_checks: include to run lint/test/typecheck/build evidence.

2. Collect evidence:

skills/evidence-heavy-evaluator/scripts/collect_evidence.sh \
  --target-dir <target_dir> \
  --profile <profile> \
  --depth <depth> \
  [--execute-checks]

3. Read outputs from <target_dir>/test-output/evidence-heavy-evaluator/:

• readiness-scorecard.json
• readiness-report.md
• checks-summary.tsv
• metrics.tsv
• signals.tsv

4. Summarize results for the user:

• Lead with highest-impact failed criteria.
• Cite the exact artifact paths used as evidence.
• Separate failed checks from skipped/not-evaluated checks.

Guardrails

• Keep evaluation read-only: do not edit code as part of this skill.
• Treat command failures as evidence, not blockers.
• Preserve deterministic ordering in report summaries.
• If --execute-checks is omitted, call out that quality execution criteria are not evaluated.

Criteria

Use references/criteria-matrix.md as the source of truth for scoring criteria and profile weights.

Notes

• The collector automatically runs render_report.py after evidence collection.
• uv is required because render_report.py is executed with uv run.