evaluator

A separate reviewer that tests code from an adversarial point of view. It checks whether the code works, matches the design, handles edge cases, and produces the claimed result rather than merely reporting success.

In plain words
What is it for?
Use it after implementation, after a sprint, or before committing code when an independent pass result is required.
Why use it?
Code can exist without actually working, and mistakes can survive if the person who wrote it is also the only person checking it.

Skill for Claude CodeCodex

Install

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.

agentmods
npx agentmods add skills/teamspwk/nova/evaluator
Any agent
npx skills add TeamSPWK/nova --skill evaluator
Clone the repo
git clone --depth 1 https://github.com/TeamSPWK/nova

Made for: Claude Code, Codex.

Per session 63 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 5,872 The whole file, excluding the scripts and references it only reads on demand.
Security scan D 3 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00063 $0.05872
Opus 5 $0.00032 $0.02936
Sonnet 5 $0.00013 $0.01174
Haiku 4.5 $0.00006 $0.00587

Measured 2d ago against content hash 9e2c063689f9, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade D, and why

evaluator scanned grade D with 3 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 2d 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.

Asks for rootmediumPrivilege escalation

A mod that escalates privileges can change anything on the machine, not only the project.

HIGH_RISK=$(jq -s '[.[] | select(.event_type=="tool_constraint_violation" and (.extra.matched_pattern | test("rm -rf|sudo |eval |dd if=|mkfs|chmod 777")))] | length' .nova/events.jsonl 2>/dev/null || echo 0)

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

HIGH_RISK=$(jq -s '[.[] | select(.event_type=="tool_constraint_violation" and (.extra.matched_pattern | test("rm -rf|sudo |eval |dd if=|mkfs|chmod 777")))] | length' .nova/events.jsonl 2>/dev/null || echo 0)

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

| **API 변경** | curl로 변경된 엔드포인트 실제 응답 확인. 상태 코드 + 응답 바디 검증 |
.claude/skills/evaluator/SKILL.md · 394 lines

How it starts

The opening of the file, as written. The whole thing — 394 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Nova Adversarial Evaluator

적용 규칙 (on-demand 로드)

  • docs/nova-rules.md §2 Generator-Evaluator 분리 + 하드 게이트 (독립 서브에이전트 기술 정의)
  • docs/nova-rules.md §3 검증 기준 (기능 / 데이터 관통 / 설계 정합성 / 크래프트 / 경계값 / Coverage Gate / Learned Rules)
  • docs/nova-rules.md §10 관찰성 계약 — 판정 직후 hooks/record-event.sh evaluator_verdict 호출

공식 용어 매핑 (Anthropic eval framework)

Anthropic — Demystifying evals for AI agents의 5대 개념과 Nova의 매핑:

공식 용어 정의 Nova 대응
task 입력 + 성공 기준을 갖춘 단일 테스트 docs/nova-rules.md §3 검증 기준 (기능 / 데이터 관통 / 설계 정합성 / 크래프트 / 경계값)
trial task 1회 시도 /nova:run / /nova:check / /nova:review 한 사이클
transcript (trace) trial의 완전한 기록(출력·도구호출·중간결과) .nova/events.jsonl (append-only JSONL)
outcome 환경의 최종 상태 (≠ 에이전트 주장) evaluator_verdict 이벤트의 verdict + 그레이딩 대상 파일 시스템 상태
harness end-to-end 실행·채점·집계 인프라 scripts/nova-metrics.sh + hooks/record-event.sh + 본 Evaluator 서브에이전트 + .nova/events.jsonl 파이프라인 전체

핵심 원칙 일치:

  • "outcome ≠ agent's self-report" → Nova 평가 자세 "코드가 존재하는 것과 동작하는 것은 다르다" (Layer 3 실행 검증 필수) 와 정합
  • "mistakes propagate and compound across turns" → Nova "재검증 프로토콜" (수정 후 자동 재검증) 의 동기
  • "trajectory metrics tell you why agents succeed or fail" → Nova tool_constraint_violation / schema_error 사후 감사 jq 쿼리가 trajectory 분석

용어 쇄신보다 병기. Nova 기존 용어(Layer 1~3·검증 기준 5종)는 그대로 유지하고, 외부 문서 크로스 레퍼런스가 필요할 때 본 매핑을 사용한다.

관찰성 훅 (v5.12.0+)

판정(PASS/CONDITIONAL/FAIL)을 내린 직후 반드시 이벤트 기록:

bash hooks/record-event.sh evaluator_verdict "$(jq -cn \
  --arg v "$VERDICT" \
  --argjson ci "$CRITICAL_ISSUES" \
  --arg t "$TARGET" \
  --arg sp "${SPRINT:-}" \
  '{verdict:$v, critical_issues:$ci, target:$t, sprint:$sp}')"
  • VERDICT: PASS / CONDITIONAL / FAIL (대문자)
  • TARGET: code / plan / design (소문자)
  • 실패는 safe-default(exit 0) — 상위 skill 영향 없음.
  • Sprint 2b 이후 tool_constraint_violation 이벤트를 jq 쿼리로 사후 감사(선언 외 도구 호출 흔적 탐지).

Read the full file on GitHub · 394 lines

Changes

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.

  1. 2d ago First seen · 394 lines · 63 tokens per session scan D 9e2c063689f9

Subscribe to this mod's changes

evaluator is a skill published in the GitHub repository TeamSPWK/nova (2 stars, last pushed 2d ago), licensed MIT. It adds 63 tokens to every session and 5,872 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it D with 3 findings (asks for root, recursive force delete, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

loopx-project

Use when connecting a repository or project goal document to LoopX, maintaining project-local goal state, refreshing stale dashboard status, syncing local projects into the shared global registry, or diagnosing LoopX CLI/PATH/status/history issues across multiple repos. For registering durable project materials such…

huangruiteng/loopx · 74 tokens

loopx-material

Operate an explicitly activated LoopX Material Lifecycle for a connected project. Use for material-store inventory, lossless migration, candidate/archive transitions, exact-read-backed ranking, ranked-entry rebuilds, bounded Explore intake, owner-gated apply, rollback, and audit. Do not use for ordinary one-off…

huangruiteng/loopx · 74 tokens

loopx-pr-program

Use when LoopX must manage a multi-PR or multi-MR delivery program across one or more repositories: inventory current change requests, reconcile new/merged/closed or retargeted work, preserve requirement and dependency priorities, maintain a roadmap document, or monitor material lifecycle/check/review changes over…

huangruiteng/loopx · 95 tokens

paperclip

Interact with the Paperclip control plane API for task coordination and governance. Use when checking assignments, updating issue status, posting comments, delegating work, managing routines, or calling Paperclip API endpoints.

paperclipai/paperclip · 43 tokens

wayfinder

Plan a huge chunk of work (more than one agent session can hold) as a shared map of decision tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.

mattpocock/skills · 46 tokens

github-labels-query

List GitHub repository labels with perpage pagination and name filtering support.

github/gh-aw · 19 tokens