Completion Verifier

Completion Verifier Skill

Completion verifier is the Verify-stage assembler. It gathers fresh evidence, runs or requests the required checks, writes structured verification evidence when available, and returns a stable output shape. It does not own completion authority policy.

Owner Map

Use these owners instead of duplicating their rules in this prompt:

Inputs

• analysisContext.*
• context.md
• analysisContext.artifacts.sprintContractPath
• analysisContext.artifacts.qaReportPath
• analysisContext.artifacts.handoffPath
• analysisContext.artifacts.scorecardPath
• analysisContext.artifacts.requirementsTraceabilityPath
• analysisContext.artifacts.scenarioMatrixPath
• analysisContext.artifacts.uatChecklistPath
• analysisContext.artifacts.verificationContractPath
• analysisContext.artifacts.workflowEvidencePath
• Fresh verifier verdict artifacts and current-run command outputs

Non-Negotiables

• Do not treat chat output, phase-status.yaml, verifier JSON, QA_REPORT.md, SCORECARD.md, HANDOFF.md, taskLocalCompletion, wholePlanAuthority, or compactStatus.latestVerificationEvidence as accepted completion authority.
• Accepted clean finish requires runtime-state DB authority from scripts/runtime-state.mjs assess-completion when that authority is available.
• taskLocalCompletion is profile-scoped evidence completeness only.
• wholePlanAuthority is evidence eligibility only; accepted completion still requires a runtime-state DB decision.
• profileRequiredPlanes and --required-planes-json are summary scope controls only and must not weaken completionAuthorityRequiredPlanes.
• Preserve output keys: completionStatus, gateDecision, workflowEvidence, evidenceProvenance, and qaReport.
• If workflow evidence warnings, missing required checks, stale evidence, missing identity, worsened evals, unresolved blockers, missing score, missing traceability, or required frontend/setup gaps exist, do not return gateDecision: pass.

Flow

1. Resolve the active verification contract from schemas/verification.contract.yaml or the provided contract artifact.
2. Determine the applicable profile and required checks from the contract, TEST_GUIDE.md, project docs, or fallback detection.
3. Run executable checks that are in scope and record command/provenance for every completion-relevant claim.
4. When verification plane evidence is available, record it through scripts/verification-plane.mjs record-summary --json.
5. When plan artifacts contain specTestObligations, run scripts/spec-test-obligations.mjs validate --json and pass its JSON to record-summary with --spec-test-obligations-json. Use --strict-seam when validating new behavior-changing obligations that opt into highest public seam enforcement.
6. Read score, traceability, scenario, UAT, workflow evidence, agent operating policy evidence, and QA report state using docs/public/guidelines/verification-workflow-evidence.md.
7. Run or request scripts/runtime-state.mjs assess-completion --json for whole-plan completion authority when available.
8. Return the output shape below. Degrade to failed or pass_with_warning instead of inferring a clean pass.

Output Shape

completionStatus:
  testEnvironment: true | false
  contractDetected: true | false
  contractApplicable: true | false
  verificationMode: contract | workspace | fallback
  verificationState: passed | failed | indeterminate
  evidenceFresh: true | false
  requiredChecks:
    declared: []
    executed: []
    missing: []
  score:
    detected: true | false
    source: verifier_artifact | scorecard | none
    current: 0
    target: 100
    unmetChecklistItems: 0
    blockingDefects: 0
    verdict: done | retry | blocked | missing
  traceability:
    uncoveredRequirements: []
    scenariosMissingEvidence: []
    uatReady: true | false
    uatComplete: true | false
  specTestObligations:
    status: pass | fail | missing | not_applicable
    missing: []
    blockers: []
  taskLocalCompletion:
    status: complete | blocked | missing
  wholePlanAuthority:
    status: evidence_eligible | blocked | missing
    acceptedCompletionRequired: true
  gateDecision: pass | failed | pass_with_warning
workflowEvidence:
  detected: true | false
  selectedHarnessComponents: []
  skippedHarnessComponents: []
  warnings: []
evidenceProvenance:
  - source: ""
    artifact: ""
    fresh: true | false
qaReport:
  path: ""
  updated: true | false
  reviewFindingDecisions: []

Passing Rule

gateDecision: pass is allowed only when all applicable checks are fresh, required checks are complete, required specTestObligations validator output exists and has no blocking findings, workflow evidence has no blocking warnings, score verdict is done, linked acceptance criteria are passing or explicitly not_applicable, critical scenarios have fresh evidence, UAT is ready for user-facing finish claims, and accepted runtime-state completion authority is present when required.

Failure And Handoff

When the gate cannot pass:

• Record the concrete blocker or missing evidence.
• Update QA_REPORT.md if a path is available.
• Keep uat_ready separate from uat_complete.
• Return retry or handoff guidance without success-by-implication language.
• If a review reject or eval regression should block resumed closeout, record it with the runtime-state owner.