validator-workflow（验证器工作流程）

Validator Phase Workflow

You have been assigned a phase to validate after a builder reports completion. Your spawn prompt contains the phase file path and plan folder. This skill teaches you how to handle validation end-to-end.

Why This Workflow Exists

The user experienced validators that ran shallow checks, missed pattern deviations, and didn't catch issues introduced by auto-fixes. Each step below prevents a specific failure:

Step 0: Load Project Rules

Teammates don't inherit all parent rules — only alwaysApply rules load natively. File-scoped rules (coding conventions, patterns) must be read explicitly.

Read these two files (they contain universal conventions for all code):

Read ~/.claude/rules/coding-style.md
Read ~/.claude/rules/patterns.md

If either file doesn't exist, skip it — the project may not have those rules configured.

These rules inform what you flag during review — import ordering, error handling, service patterns, data fetching conventions.

Step 1: Create Task List

Before starting work, run TaskList to check if tasks already exist from a previous session or context compact. If tasks exist with your agent name as owner, resume from the first in_progress or pending task.

If no tasks exist, create them now. Prefix all task subjects with [Review] to distinguish from builder and orchestrator tasks. Always set owner to your agent name.

TaskCreate({
  subject: "[Review] Run code review for Phase {NN}",
  description: "Invoke /code-review against phase file. Write review to {plan-folder}/reviews/code/phase-{NN}.md",
  activeForm: "Running code review",
  metadata: {
    created_by: "{your-agent-name}",
    agent_type: "validator",
    phase: "P{NN}",
    group: "{group-name}",
    skill: "{skill-from-phase}",
    role: "review",
    attempt: 1,
    parent_task_id: "{orchestrator-phase-task-id-from-spawn-prompt}"
  }
})

After creating, set yourself as owner:

TaskUpdate({ taskId: "{id}", owner: "{your-agent-name}" })

Standard validator tasks:

Mark each task in_progress before starting and completed when done.

Step 2: Read the Phase

Read the phase file from your spawn prompt. Extract:

• skill: field from frontmatter — determines which extra tests to run (Step 3)
• Acceptance criteria — your success metrics for the verdict
• Implementation steps — what was supposed to be built
• Files created/modified — scope of your review

Step 3: Run Code Review

Invoke the code review skill against the phase:

Skill({ skill: "code-review", args: "[phase-file-path]" })

This forks a sub-agent that:

• Reads the phase document and extracts all implementation steps
• Finds reference implementations from the codebase (ground truth)
• Reviews each file against phase spec AND codebase patterns
• Auto-fixes Critical/High/Medium issues directly in source files
• Writes a review file to {plan-folder}/reviews/code/phase-{NN}.md
• Returns a verdict with issue counts and what was fixed

Step 4: Run Verification

Skip verification entirely if the code review verdict is "Ready" with zero auto-fixes — the builder already passed tests + typecheck before reporting, and no source files changed since.

Run verification if the code review auto-fixed any issues (files were modified):

Always Run

pnpm run typecheck
pnpm test

Both must pass. If auto-fixes introduced issues, fix them.

Conditional Testing by Phase Type

The phase's skill: frontmatter field determines which additional tests to run:

E2E Test Scoping

When running E2E tests, scope them to the feature being validated:

1. Extract keywords from the phase title/slug (e.g., notes, billing, auth)
2. Glob for e2e/**/*{keyword}*.spec.ts
3. If matches found: pnpm test:e2e -- [matched spec files]
4. If no matches: pnpm test:e2e (full suite)

Graceful Skip

If a test command doesn't exist (exit code from missing script), skip it and note in the report:

E2E tests: skipped (pnpm test:e2e not configured)

This allows projects without E2E or DB tests to pass validation without false failures.

Step 5: Determine Verdict

Based on the code review results and verification (if it ran):

• PASS: Code review verdict is "Ready", no unfixed Critical/High issues, and all verification passed (or was skipped because no files changed)
• FAIL: Any unfixed Critical/High issues, or (if verification ran) typecheck errors, test failures, E2E failures, or DB test failures

Step 6: Report to Orchestrator

SendMessage({
  type: "message",
  recipient: "team-lead",
  content: "Phase [NN] validation: [PASS|FAIL]\n\nCode review: [verdict]\nReview file: [path]\nVerification: [pass|skipped (no changes)]\nE2E tests: [pass|fail|skipped|N/A]\nDB tests: [pass|fail|skipped|N/A]\n\n[If FAIL: specific issues with file:line references and exact fixes needed]",
  summary: "Phase NN: PASS|FAIL"
})

Step 7: Go Idle

Wait for the next validation assignment or a shutdown request.

FAIL Reports Must Be Actionable

When reporting FAIL, include enough detail for a fresh builder to fix the issues without guessing:

• File:line references for each issue
• Which pattern was violated (cite the reference file)
• Exact fix needed (not "consider improving" — state what must change)
• Which test failed (test name, assertion, expected vs actual)

Vague FAIL reports cause fix builders to guess, producing more failures. Specific reports enable one-shot fixes.

IMPORTANT: Before using the Write tool on any existing file, you MUST Read it first or the write will silently fail. Prefer Edit for modifying existing files.

Resuming After Context Compact

If your context was compacted mid-validation:

1. TaskList → scan for tasks where you are the owner (your agent name)
2. Find your in_progress task, or if none, your first pending task
3. TaskGet on that task → read the description AND metadata
4. Metadata tells you: which phase (phase), which group (group), and which orchestrator task you report to (parent_task_id)
5. Continue from that task — don't restart the validation
6. The task list and metadata are your source of truth, not your memory