uat-verify

name	uat-verify
description	Verify completed work against acceptance criteria, collect evidence, and produce a PASS/FAIL UAT report
version	1.0.0
tools	["Read","Bash","Grep","Glob","TaskUpdate"]
source	builtin
trust_score	100
provenance_sha	be9f7b3ee10b09d8

UAT Verify Skill

Purpose

Systematically verify that a completed feature or task satisfies every acceptance criterion before marking work done. Produces a structured PASS/FAIL report with concrete evidence.

When to Invoke

Skill({ skill: 'uat-verify' })

Invoke when:

• A developer or agent has marked work complete and you need to independently verify it
• A task has acceptance criteria that must be validated before closing
• A pull request or deliverable needs UAT sign-off
• You need to produce an evidence-backed completion report

---

Workflow

Step 1: Extract Acceptance Criteria

Read the task, plan, or user story:

# Read task details
TaskGet({ taskId: '<id>' })
# or read plan file
Read('.claude/context/plans/<feature>-plan.md')

Extract all acceptance criteria into a numbered list. Look for:

• Bullet points under "Acceptance Criteria", "Definition of Done", "Requirements"
• "must", "should", "expected to", "given/when/then" patterns
• Explicit test assertions or feature descriptions

Step 2: Collect Evidence Per Criterion

For each criterion, execute verification steps:

# Run automated tests
node --test tests/path/to/feature.test.cjs

# Execute feature and capture output
Bash({ command: '<command that exercises the feature>' })

# Inspect implementation
Read('<path/to/implementation/file>')

# Search for required patterns
Grep({ pattern: '<pattern>', path: '<directory>' })

Tag every output to the criterion it evidences.

Step 3: Score Each Criterion PASS or FAIL

For each criterion:

• PASS — evidence directly satisfies the criterion statement. Cite: test name, line output, or file reference.
• FAIL — evidence is missing, contradicts the criterion, or the feature does not behave as specified. Document the gap.

Step 4: Produce Verdict

• All criteria PASS → overall PASS
• Any criterion FAIL → overall FAIL

Step 5: Write Results Report

Use the UAT results template:

Read('.claude/templates/uat-results.md')

Fill in all placeholders. Write the completed report to: .claude/context/reports/backend/uat-{feature}-{YYYY-MM-DD}.md

Then call:

TaskUpdate({
  taskId: '<id>',
  status: 'completed',
  metadata: {
    summary: 'UAT complete: PASS/FAIL — N/M criteria satisfied',
    reportPath: '.claude/context/reports/backend/uat-{feature}-{YYYY-MM-DD}.md',
  },
});

---

Anti-Patterns

• Never claim PASS without executing the feature — reading code is not proof it works
• Never skip failing criteria in the report — all FAILs must be documented
• Never produce a "partial PASS" — the verdict is binary
• Never run test execution inside a read-only review — use Bash to actually execute

Related

• .claude/workflows/enterprise/verify-phase-uat.md — detailed 5-phase workflow
• .claude/templates/uat-results.md — report template