Paths: File paths (shared/ , references/ , ../ln-* ) are relative to skills repo root.

Type: L2 Coordinator Category: 5XX Quality

Quality Coordinator

Evaluation-platform coordinator for story quality review.

Mandatory Read

MANDATORY READ: Load shared/references/evaluation_coordinator_runtime_contract.md , shared/references/evaluation_summary_contract.md , shared/references/evaluation_research_contract.md , shared/references/loop_health_contract.md

MANDATORY READ: Load shared/references/agent_review_workflow.md , shared/references/agent_delegation_pattern.md

MANDATORY READ: Load references/criteria_validation.md , references/gate_levels.md

Purpose

• invoke ln-511-code-quality-checker

• invoke ln-512-tech-debt-cleaner

• invoke ln-513-regression-checker

• invoke ln-514-test-log-analyzer

• run inline agent review in parallel with read-only evidence gathering

• keep merge, refinement, and verdict sequential

• return normalized quality results

Inputs

Primary input:

• storyId

• --previous-cycle-focus (optional, from ln-500): comma-separated blocking categories from prior FAIL cycle

Status filter:

• To Review

Critical Rule

Fast-track paths that skip research are not allowed.

Every quality run must include:

• official documentation or standards

• MCP Ref

• Context7 when a framework or library is involved

• current web best-practice research

Runtime Contract

Runtime family:

• evaluation-runtime

Identifier:

• quality-{storyId}

Phase order:

• PHASE_0_CONFIG

• PHASE_1_DISCOVERY

• PHASE_2_READ_ONLY_EVIDENCE

• PHASE_3_CLEANUP

• PHASE_4_AGENT_BARRIER

• PHASE_5_MERGE

• PHASE_7_REFINEMENT

• PHASE_8_VERDICT

• PHASE_9_SELF_CHECK

Worker Invocation (MANDATORY)

Use the Skill tool for delegated workers. Do not inline worker logic inside the coordinator.

TodoWrite format (mandatory):

• Resolve Story and build runtime manifest

• Load Story metadata and detect changed files

• Run quality checkers and research in parallel

• Apply safe tech-debt cleanup

• Sync agents and wait for all evidence

• Merge and deduplicate all findings

• Run bounded refinement loop

• Compute quality verdict and score

• Verify runtime cleanup and self-check

Representative invocations:

Skill(skill: "ln-311-review-research-worker", args: "{storyId} quality research") Skill(skill: "ln-511-code-quality-checker", args: "{storyId}") Skill(skill: "ln-512-tech-debt-cleaner", args: "{storyId}") Skill(skill: "ln-513-regression-checker", args: "{storyId}") Skill(skill: "ln-514-test-log-analyzer", args: "{storyId}")

Workflow

Phase 0: Config

• Resolve storyId .

• Build evaluation runtime manifest with required_research=true .

• Start evaluation-runtime .

Phase 1: Discovery

• Load Story metadata and completed implementation task scope.

• Detect changed files and project stack.

• Index semantic graph when available.

Phase 2: Read-Only Evidence

Parallel work allowed in this phase:

• ln-311-review-research-worker

• ln-511-code-quality-checker

• ln-513-regression-checker

• ln-514-test-log-analyzer

• external agent launch

Rules:

• research is mandatory

• worker summaries are the only completion signal

• no merge or mutation occurs in this phase

When previous_cycle_focus is provided:

• Prioritize evidence collection for the listed blocking categories.

• ln-511 code quality checker should focus on the specified areas first.

• This does not exclude other evidence — it reorders priority.

Phase 3: Cleanup

• Run ln-512-tech-debt-cleaner only after read-only evidence is collected.

• Cleanup remains sequential because it mutates files.

• Record the worker summary and any cleanup evidence.

Phase 4: Agent Barrier

• Sync agents through evaluation-runtime .

• Do not cross this barrier until all required agents are resolved or explicitly skipped.

• Treat failure_class from agent results as transport evidence:

• rate_limited , tool_missing , auth_missing , permission_denial , and asked_question are not quality FAIL findings by themselves.

• timeout_productive can continue to merge/review only when output/log/session evidence exists.

• repeated identical worker/agent failure without new artifacts pauses through loop health before another cycle.

Phase 5: Merge

Merge inputs:

• research summary

• ln-511 summary

• ln-512 summary

• ln-513 summary

• ln-514 summary

• agent findings

Rules:

• deduplicate before scoring

• unsupported claims are rejected

• security and correctness issues remain high priority

Phase 6: Refinement

Refinement uses a 2-stage state machine (see ln-316-review-refinement-worker ):

• Stage 1 (parallel): dry_run_executor , new_dev_tester , adversarial_reviewer

• Stage 2 (after merge): final_sweep

Rules:

• Stage 1 runs in parallel, Stage 2 after merge

• cleanup evidence required for spawned processes

• no research skipping

Phase 7: Verdict

Compute normalized quality verdict using:

• code quality

• cleanup result

• agent review

• criteria validation

• linter result

• regression result

• log analysis result

Final verdict values:

• PASS

• CONCERNS

• FAIL

Phase 8: Self-Check

Required checks:

• runtime started

• mandatory research completed

• all worker summaries recorded

• all required agents resolved before merge

• cleanup verified

• refinement trace recorded when applicable

• coordinator summary written

Summary Contract

Write summary_kind=evaluation-coordinator .

Recommended payload:

• status

• final_result

• report_path

• worker_count

• agent_count

• issues_total

• severity_counts

• warnings

• cleanup_verified

• research_completed

Definition of Done

• Evaluation runtime started

• Mandatory research completed

• Read-only evidence workers completed

• Cleanup worker completed or justified

• Agent barrier resolved

• Merge completed

• Refinement executed or explicitly justified

• Final verdict calculated

• evaluation-coordinator summary written

• Runtime completed

Meta-Analysis

MANDATORY READ: Load shared/references/meta_analysis_protocol.md

After the coordinator run, analyze the session per protocol section 7 and include the protocol-formatted output with the final quality verdict.

References

• Runtime: shared/references/evaluation_coordinator_runtime_contract.md , shared/references/evaluation_summary_contract.md

• Research: shared/references/evaluation_research_contract.md

• Workers: ../ln-511-code-quality-checker/SKILL.md , ../ln-512-tech-debt-cleaner/SKILL.md , ../ln-513-regression-checker/SKILL.md , ../ln-514-test-log-analyzer/SKILL.md

• Quality criteria: references/criteria_validation.md , references/gate_levels.md

Version: 7.0.0 Last Updated: 2026-02-09