Harnie Reviewer — Development agent for Claude Code
Read-only code reviewer with a REJECT bias.
How to install Harnie Reviewer
Installs to ~/.claude/agents/qnamy-harnie-harnie-reviewer.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/qnamy/harnie/HEAD/agents/harnie-reviewer.md -o ~/.claude/agents/qnamy-harnie-harnie-reviewer.md Restart Claude Code, or start a new session, for it to be picked up.
What Harnie Reviewer does
name: harnie-reviewer description: Read-only code reviewer with a REJECT bias. Reviews the builder's changes in the cross-model code loop and returns the review-schema.md VERDICT/ISSUES schema. Never creates or modifies files. tools: Read, Grep, Glob model: opus
You are a **read-only code reviewer**. Review changes from the producer, the builder, in the cross-model build loop. **You are not the producer.** As the builder's opposite-provider reviewer, your role is to reduce cross-model b
Alternatives in Development
- Eval Implementer — Implementer role of the D365FO agent eval loop 136 ★
- Re Doc Organizer — PHẢI dùng SAU KHI re-lead đã duyệt (VERDICT=ĐẠT/CORRECT) kết quả reverse-engineering 1 model/tính năng — viết 90 ★
- Context Builder — Builds architectural understanding before any verdict — runs Phase 0 setup and Phase 0.5 Context Reconstructio 50 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Harnie Reviewer Ko
REJECT 편향의 읽기 전용 코드 리뷰어. 크로스-모델 코드 루프에서 빌더의 변경을 리뷰하고 review-schema.md의 VERDICT/ISSUES 스키마를 반환한다. 파일을 생성·수정하지 않
Light Reviewer
Read-only reviewer for routine diffs, bug fixes, and regression checks. Reports concrete correctness issues an
Goal Verifier Agent
Goal-level checker for the /goal loop: evaluates a goal's completion_criteria against repo reality and returns
Self Review Verifier
Read-only adversarial verifier for the self-review loop. Takes one or more candidate findings plus the scope f
Loop Judge
Fresh-context cold-read judge for subjective acceptance criteria (readability, API clarity). Runs after mechan
Convergence Monitor
Cross-iteration convergence tracker — runs 6 detection algorithms (score-based, semantic, diminishing returns,
Related Skills
Helpful Content Review Skill
A Claude skill that reviews and scores any draft or live page against Google's helpful, reliable, people-first
Babysit PR
Babysit a single PR — fix CI issues, address review comments. Bias strongly toward fixing. Escalation is a las
Verify Task
Optional independent check of a built ticket. Reviews the PR against the Jira ticket and the approved plan, ru