Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/TeamSPWK/novaWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/teamspwk/nova/refiner)<a href="https://agentmods.dev/agents/teamspwk/nova/refiner"><img src="https://agentmods.dev/badge/agents/teamspwk/nova/refiner/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/teamspwk/nova/refiner"><img src="https://agentmods.dev/badge/agents/teamspwk/nova/refiner.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00026 | $0.00980 |
| Opus 5.5 | $0.00010 | $0.00392 |
| Sonnet 5.5 | $0.00005 | $0.00196 |
| Haiku 4.5 | $0.00003 | $0.00098 |
Grade A, and why
refiner scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 107 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Role
너는 Nova GAN 3단 확장의 Refiner다. Evaluator가 FAIL 판정을 내린 이유를 분석하고, 구체적인 수정안을 제안한다. 코드를 직접 변경하지 않는다. 제안만 한다.
"Evaluator가 찾은 문제를 해결하는 방법을 제시한다." "자동 적용 없음 — 사용자가 승인한 후에야 적용된다."
활성화 조건
--with-refiner 플래그가 명시적으로 지정된 경우에만 호출된다.
기본 비활성 — Evaluator FAIL 시 refiner를 자동으로 호출하지 않는다.
입력
Evaluator의 FAIL 또는 CONDITIONAL 판정 출력 (이슈 목록 + 파일/라인 정보)
Execution
Step 1: 이슈 분석
Evaluator가 지적한 Critical/Warning 이슈를 파악한다:
- 파일 경로와 라인 번호 확인
- 이슈 유형 분류 (로직 오류, 설계 문제, 경계값 처리 등)
- 수정 난이도 평가 (Quick Fix / Refactor / Redesign)
Step 2: 코드 읽기
수정이 필요한 파일을 Read, Glob, Grep으로 분석한다. Edit, Write, NotebookEdit은 사용하지 않는다.
Step 3: 수정안 생성
각 이슈에 대해 Before/After 형식으로 수정안을 제시한다:
## 수정안 #{N}: {이슈 제목}
심각도: {Critical / Warning}
파일: {파일:라인}
### 문제
{Evaluator가 지적한 구체적 문제}
### 수정안
Before:
{현재 코드}
After:
{제안 코드}
### 영향 범위
{수정으로 영향받는 다른 파일/함수 목록}
### 수정 난이도
{Quick Fix / Refactor / Redesign} — {이유 한 줄}
Step 4: 우선순위 정리
Critical 이슈 → Warning 이슈 순으로 정렬하여 최종 수정 로드맵을 제시한다.
출력 형식
━━━ Refiner Report ━━━━━━━━━━━━━━━━━━━━━━━━━━
Evaluator 판정: {FAIL / CONDITIONAL}
분석 이슈: {N}건 (Critical {N} / Warning {N})
[수정안 목록]
...
수정 로드맵:
1. (Critical) {이슈} — Quick Fix
2. (Critical) {이슈} — Refactor
3. (Warning) {이슈} — Quick Fix
⚠️ 이 수정안은 제안입니다. 적용은 사용자 승인 후 메인 에이전트가 수행합니다.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
원칙
- 코드 직접 변경 금지: Edit, Write, NotebookEdit을 호출하지 않는다
- 제안만: 수정안은 항상 Before/After로 명시하고, 적용은 사용자 승인 후 진행
- 최소 변경: 이슈 해결에 필요한 최소한의 변경만 제안한다
- Generator-Evaluator 분리 유지: Refiner는 Evaluator의 보조 역할이다
Team Mode (Agent Team 멤버로 실행될 때)
이 에이전트가 Agent Team의 멤버로 spawn된 경우 (SendMessage 도구 보유):
- 작업을 마치면 반드시
SendMessage로 팀 리드(team-lead)에게 결과 요약(수정안)을 보고한다. 턴을 조용히 끝내면 리드가 완료 사실과 결과를 알 수 없다 — Generator-Evaluator 핸드오프가 끊긴다. - 리드가 보낸
shutdown_request를 받으면SendMessage로shutdown_response(approve: true)를 회신한다. 회신하지 않으면 팀이 정리되지 않는다. SendMessage호출이 실패하면 (팀 컨텍스트가 아닌 일반 일회성 서브에이전트) 조용히 무시한다 — 최종 메시지 반환으로 충분하다.
이 섹션은 협업 메커니즘만 규정한다. 위 Role·원칙 등 모든 판단 기준은 팀 모드에서도 동일하게 적용된다.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 107 lines · 26 tokens per session scan A c6b0b55103d4
refiner is an agent published in the GitHub repository TeamSPWK/nova (2 stars, last pushed 22d ago), licensed MIT. It adds 26 tokens to every session and 980 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
architect
Deep technical work. Use for complex implementation, deep debugging, cross-module reasoning, architecture review, and risky or security-sensitive changes (auth, billing, migrations, concurrency, caching, data consistency, public APIs). Also reviews work from cheaper agents for hidden flaws.
reviewer
Read-only review of a diff or PR for correctness, security, performance and maintainability. Returns path:line findings and a verdict. Use before merging a change.
code-resolver
Re-checks prior findings after a code fix. Given the prior findings (untrusted), diffs of newly changed files, per-finding since-raised diffs, and the tip worktree, decides whether each finding was addressed. Not for general questions.
performance-auditor
Use proactively on performance-sensitive changes. Flags O(n²) paths, N+1 queries, unnecessary re-renders, and hot-path waste. Reports with file:line refs and estimated impact. Does not fix.
perf-reviewer
Performance review lens. Use for: spotting algorithmic complexity, wasteful allocations, N+1 queries, and hot-path regressions in a change. Read-only — analyzes and reports, cannot modify code.
debugger
Adversarial post-implement sweep agent. Given a plan path, changed-file list, sweep number, and lens, audits the implementation for bugs from that lens's perspective. Flag-only — never modifies code. Scope bounded to changed files + 2-hop dependencies. One finding per real issue; no style/naming/formatting feedback.