listener-he/java-harness-agent
14 subagents
Subagent
ambiguity-gatekeeperGATE on ambiguous input — enforce definition-of-ready (clear scope + testable outcome + explicit AC) before AC transcription starts.code-reviewerREVIEW newly written code (the diff) for correctness, performance, security, and maintainability — fresh-context inspection in an isolated sub-agent.database-reviewerReview changes to MyBatis mapper XML, `*Mapper.java`, SQL migrations, or DDL/DML for project-specific anti-patterns. TRIGGER when an Edit/Write touches `src/main/resources/mapper/**/*.xml`,…documentation-curatorAUTHOR documentation grounded in real source — README, API/Javadoc, migration guide, runbook, architecture writeup, ADR explainer. Every claim traceable to a file path or commit; never invents names,…java-build-resolverDiagnose Java/Maven build failures and propose concrete fixes. TRIGGER when `mvn compile` / `mvn test-compile` / `javac` fails during Phase 4 Implement.knowledge-architectSPLIT oversized wiki index files into focused sub-documents and rewrite the original as a lean routing graph. TRIGGER when a wiki index (`.claude/wiki/wiki/*/index.md`) exceeds the 3000-line cap (per…knowledge-extractorEXTRACT stable knowledge from completed code changes into structured WAL fragments. Writes ONLY the dimensions (Domain, API, Rules, Data, Architecture) the user elected via `h-archive` Step 3b —…lead-engineerIMPLEMENT code per `task_brief.md` Machine Section — translate Allowed Scope + ACs (Given/When/Then) + Hard Constraints into concrete, compilable Java/Maven changes following project standards.librarianMAINTAIN wiki health via two flows. **Compact**: merge scattered WAL fragments into stable index files + garbage-collect merged fragments. **Distill**: scan for stale/duplicate knowledge files,…requirement-engineerAC TRANSCRIPTION ENGINE for non-PRD inputs. Converts a raw Idea / Feedback / Compliance / Security ask into testable Given/When/Then Acceptance Criteria plus a structured Must-Ask question list for…security-sentinelSCAN for secret leakage + authorization-bypass risks via deterministic scripts — pure tool runner, no subjective security review.system-architectDESIGN system architecture BEFORE any code is written — high-level interactions, database schema, API contracts, design patterns, and irreversible decisions captured as ADRs. Acts as Foreman in EPIC:…test-runnerRun JUnit/Surefire tests scoped to changed modules, parse output, and return a structured `AC-id → test method → PASS/FAIL/SKIP` mapping. TRIGGER at Phase 5 QA, especially when AC count ≥ 4 OR risk =…triage-reviewerSEMANTIC SECOND-OPINION on prompts where the main agent genuinely cannot disambiguate intent on a HIGH-sensitivity surface. EXPLICIT dispatch only — the agent decides when to call (no auto-trigger…