scorecard
Invoke as $cc-scorecard. The user request follows the skill mention.
Scorecard Evaluation Workflow
Scope: $ARGUMENTS
Step 0 — CCEP Bootstrap
Command: scorecard (fixed for this workflow — do not infer from user text)
- Run:
npx cc-codeconductor ccep profile scorecard --output json to get the phases and their roles.
- For each delegated phase, run:
npx cc-codeconductor ccep compile --command scorecard --phase <phase-id> "$ARGUMENTS" --view prompt --output json
- After planner/intake JSON is available, run:
npx cc-codeconductor ccep evaluate --command scorecard --input <planner.json> --output json. If stop is true, show questions or risks and wait for human input.
- Pass each subagent only the compiled
prompt for its phase — never forward raw $ARGUMENTS to planners.
Step 1 — Identify task
Use $ARGUMENTS as task id (e.g. BC-001) or read active item from npx cc-codeconductor openspec status.
If a change folder exists, run npx cc-codeconductor openspec analyze --output json first. --from-diff overlays FR/SC coverage onto acceptance and TDD evidence onto tests.
Step 2 — Create scorecard with auto-signals
npx cc-codeconductor scorecard create --task <id> --agent reviewer --from-diff
Review auto-suggested criteria (minimal diff, cc-gain). Complete remaining scores 0–3 per docs/agent-scorecard.md.
Step 3 — Regression (optional)
npx cc-codeconductor scorecard regression
If required checks fail, STOP and report failures.
Step 4 — Record outcome
npx cc-codeconductor scorecard record --task <id> --agent reviewer --model <model> --verdict PASS --score 2.5
Include --cost and --tokens when available from session metrics.
Step 5 — Aggregate
npx cc-codeconductor scorecard aggregate
Report pass rate and average weighted score.
Routing on verdict
- PASS — accept deliverable; update backlog if applicable
- REVISE — return to implementer/tester with findings
- REJECT — route to task-coach for re-scoping
Apply skill evaluation.