/evaluate

Evaluate review gates and create execution eval artifacts.

Process

1. Find the execution plan. Read .claude/state/active-executions/*.yaml to find the active execution plan ID. Read the plan from artifacts/execution-plans/.
2. Collect required gates. Extract spec.required_inputs.review_gates from the plan. If empty, print "No review gates defined. Gate evaluation not required — skipping." and exit.
3. Read each gate. For each review gate ID, read the artifact to get:
   • metadata.gate_class (blocking, warning, manual-review, escalation)
   • spec.evaluation_method (static_check, human_review, diff_analysis, etc.)
   • spec.pass_condition (what must be true to pass)
4. Evaluate each gate. Based on evaluation_method:
   • diff_analysis: Check the git diff against the plan's in_scope_paths and out_of_scope_paths
   • static_check: Run available linters/checkers, or analyze code for the pass_condition
   • spec_coverage: Check that all behavior specs have traced implementations
   • human_review: Ask the user via AskUserQuestion to provide their assessment
   • contract_check: Verify contract artifacts have matching implementations
   • other: Ask the user to evaluate manually
5. Find the diff artifact. Look for a generated_diff artifact in artifacts/diffs/ that traces to the execution plan (check trace.upstream for the plan ID). If none exists (e.g., running /evaluate standalone without /run), use the plan ID as subject_diff and note "No generated_diff artifact found — evaluation based on git diff" in the evidence.
6. Create execution eval artifacts. For each gate, write:

kind: execution_eval
schema_version: "0.3"
id: EE-<short-slug>-<current ISO-8601 timestamp>
title: "Gate evaluation: <gate title>"
status: completed
owners:
  - <current user>
created_at: <current ISO-8601 timestamp>
updated_at: <current ISO-8601 timestamp>
trace:
  upstream:
    - target_id: <execution_plan_id>
      edge: depends_on
    - target_id: <review_gate_id>
      edge: depends_on
    - target_id: <generated_diff_id>
      edge: evaluates
  downstream: []
approvals: []
metadata:
  eval_type: gate_check
spec:
  subject_diff: <generated_diff_id>
  subject_gate: <review_gate_id>
  result: <pass|warn|fail|escalate>
  findings:
    - <issue description, or empty if pass>
  evidence:
    - <evidence description>

7. Write to disk. Save each to artifacts/execution-evals/<id>.yaml
8. Report summary:

Gate Evaluation Summary
=======================
Review gates: N total
  pass: N
  warn: N
  fail: N

Results:
  RG-xxx "Gate title" → PASS
  RG-xxx "Gate title" → FAIL (blocking)

9. Warn on blocking failures. If any gate with gate_class: blocking has result fail, warn: "BDSK-GATE-001: Blocking gate <id> failed. /accept will compute 'rejected' unless a waiver is approved."

Rules

• If no review gates are defined, exit cleanly (not an error)
• For human_review gates, ALWAYS ask the user — never auto-pass
• For blocking gates that fail, explicitly warn about acceptance impact
• Never modify the review gate artifacts — only read them
• Never create waiver artifacts — that is a separate governance action

Schema Compliance

• spec allows exactly 5 fields: subject_diff, subject_gate (optional), result, findings, evidence. Do NOT add description, gate_title, gate_class, summary, or any extra field. Schema uses additionalProperties: false.
• metadata allows exactly 1 field: eval_type. No additional metadata fields. Schema uses additionalProperties: false.
• Trace refs MUST be {target_id: <id>, edge: <edge>} objects, not bare string IDs.
• Approvals MUST use {authority_role: <role>, approver: <user>, approved_at: <ISO-8601>}. Not {date, decision}.
• Valid trace edges (10 total): depends_on, derived_from, constrains, implements, proves, evaluates, produced_by, supersedes, escalates_to, waives.
• Valid authority roles (5 total): product_authority, technical_authority, security_authority, release_authority, qa_authority.

Edge-kind rules for execution_eval

• Upstream depends_on → execution_plan, review_gate, execution_log
• Upstream evaluates → generated_diff (ONLY — not execution_plan or review_gate)
• Do NOT use evaluates for the link to execution_plan or review_gate — use depends_on

  # WRONG:
  upstream:
    - target_id: EP-feature-001
      edge: evaluates
  # RIGHT:
  upstream:
    - target_id: EP-feature-001
      edge: depends_on
    - target_id: GD-feature-001
      edge: evaluates