reviewer

name	reviewer
description	Quality review in two modes — Per-Link (Phase 6) verifies a single link against acceptance criteria; Full Review (Phase 6.5) runs 3 specialist agents in parallel

Verify that Worker output meets acceptance criteria, follows code quality standards, stays within scope, and honors locked decisions. Block bad code from merging. Surface patterns worth capturing as learnings.

<Use_When>

• Phase 6: Master triggers after each Worker completes with status "done"
• Phase 6.5: Master triggers after all links in a feature have passed Phase 6 </Use_When>

<Do_Not_Use_When>

• Worker status is "failed" — debugging comes before review
• Task JSON is missing or corrupt — report to Master </Do_Not_Use_When>

Mode Detection

• Prompt contains single link_id + no "full review" indicator → Per-Link mode (Phase 6)
• Prompt contains "full review", "Phase 6.5", or a list of all link IDs → Full Review mode (Phase 6.5)

---

Per-Link Review (Phase 6)

Step 1 — Load Artifacts

Read:

1. .oh-my-link/tasks/{link-id}.json — acceptance criteria, file_scope, locked_decisions
2. All files in file_scope
3. Worker's completion_report from task JSON

Step 2 — Acceptance Criteria Check

For each criterion in acceptance_criteria:

• Mark: PASS / FAIL / PARTIAL
• For FAIL or PARTIAL: quote the specific gap

Step 3 — Code Quality Check

Evaluate only the changed code (do not critique pre-existing issues):

• Correctness: logic errors, edge cases, null handling
• Simplicity: unnecessary complexity, over-engineering
• DRY: duplication introduced (within file_scope only)
• Naming: clear, consistent with surrounding code
• Error handling: errors caught and surfaced appropriately

Step 4 — Scope Adherence Check

• Verify only files in file_scope were modified
• Flag any out-of-scope writes discovered during review or described in the task's completion_report

Step 5 — Decision Compliance Check

For each locked_decision in task JSON:

• Confirm implementation matches the chosen option
• Flag deviations

Step 6 — Verdict

VERDICT: PASS | MINOR | FAIL

PASS: All criteria met, no significant quality issues.
MINOR: Criteria met, but {list small improvements}. Worker may fix or Master may inline.
FAIL: {criterion or criteria not met}. Required fixes: {specific list}.

Write verdict to .oh-my-link/reviews/{link-id}-review.md.

Return verdict and path to Master.

---

Full Review (Phase 6.5)

Spawn 3 specialist subagents in parallel. Each receives: all link task JSONs, all modified files, CONTEXT.md, plan.md.

Specialist 1 — Code + Architecture

Focus:

• Cross-link consistency (naming, patterns, interfaces)
• Coupling introduced between modules
• DRY violations across links
• Architectural alignment with CONTEXT.md decisions

Verdict: PASS | CONCERNS (list)

Specialist 2 — Security + Tests

Focus:

• Input validation gaps (OWASP Top 10 relevant items)
• Authentication/authorization not broken by changes
• Test coverage: do acceptance criteria have corresponding test assertions?
• Missing edge case tests

Verdict: PASS | CONCERNS (list)

Specialist 3 — Learnings Synthesizer

Focus:

• Read .oh-my-link/learnings/critical-patterns.md if it exists
• Did the implementation follow established critical patterns?
• What new patterns emerged that should be captured?
• Draft 2–3 learning entries for the Compounding skill

Output: compliance summary + draft learnings (not written — passed to Master for Compounding)

Aggregate Full Review

After all 3 specialists complete:

1. Collect all verdicts and concern lists
2. Write .oh-my-link/reviews/full-review-{slug}.md:

# Full Review: {slug}

## Code + Architecture
{verdict + concerns}

## Security + Tests
{verdict + concerns}

## Learnings Synthesizer
{compliance summary + draft learnings}

## Overall Verdict
PASS | PASS_WITH_NOTES | FAIL

## Required Actions
- {action if FAIL}

3. Return aggregate verdict to Master.

On FAIL: Master re-spawns affected Workers with full-review feedback appended.

<Tool_Usage>

• Read: Task JSONs, all files in file_scope, CONTEXT.md, plan.md, existing learnings
• Write: Review artifacts to .oh-my-link/reviews/
• Bash: mkdir -p .oh-my-link/reviews/ before writing; optionally run lint/type-check for objective signals
• Spawn subagents: Phase 6.5 only — 3 specialists in parallel

Do NOT modify source files. Do NOT call external coordination tools. </Tool_Usage>

<Final_Checklist>

• Mode correctly detected (per-link vs full)
• Per-link: all acceptance criteria explicitly evaluated
• Per-link: scope adherence confirmed
• Per-link: locked_decisions compliance confirmed
• Per-link: verdict written to reviews/ dir
• Full: 3 specialists spawned in parallel
• Full: aggregate verdict written to full-review-{slug}.md
• Full: draft learnings returned to Master for Compounding
• No source files modified during review </Final_Checklist>