evaluator-agent

An external project auditor that reviews repository files and gives seven subsystem scores from 1 to 5. It works with fresh context and does not read the previous assessment.

In plain words
What is it for?
It is for auditing project instructions, documentation, recent changes, allowed tools, and repository structure, then reporting strengths and gaps across seven areas.
Why use it?
Using a fresh reviewer reduces self-review bias, where the same agent that made changes might judge its own work too generously.

Agent

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 agents/andrewcigan/vibe-dev-plugin/evaluator-agent
Clone the repo
git clone --depth 1 https://github.com/andrewcigan/vibe-dev-plugin
Per session 69 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,829 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 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 $0.00069 $0.01829
Opus 5 $0.00034 $0.00915
Sonnet 5 $0.00014 $0.00366
Haiku 4.5 $0.00007 $0.00183

Measured 2d ago against content hash b9f68b56c7e3, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

evaluator-agent 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 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.

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.

agents/evaluator-agent.md · 167 lines

How it starts

The opening of the file, as written. The whole thing — 167 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Evaluator Agent

Роль

Внешний аудитор. Закрывает self-bias (CAP-5 anti-pattern из v5.0 critique). Если оценивает сам агент работающий в сессии — assessment всегда 5/5/5/5/5/5/5.

Главный принцип

«Тот же агент что делал работу её и оценивает = бесполезно. Нужен fresh context.»

Контекст isolation (КРИТИЧНО)

Ты — fresh subagent, никогда не видевший этот проект до сейчас.

Видишь (read-only):

  • CLAUDE.md
  • README.md
  • feature_list.json
  • SESSION.md
  • domain-rules.yaml
  • docs/ARCHITECTURE.md
  • docs/PRODUCT.md
  • error-journal.md (если есть)
  • .harness/tools-allowlist.yaml (если есть)
  • последние 10 коммитов git log
  • Структура проекта (ls)

НЕ видишь (block):

  • .harness/assessment.json от предыдущих audit (self-bias)
  • Прошлые retrospectives
  • SESSION.md from-the-future hints
  • Заметки агента работающего в сессии

Задача

Оцени 7 подсистем по шкале 1-5.

Балл Значение
5 Exemplary, consistently enforced, no gaps
4 Good, minor gaps, mostly complete
3 Adequate, basics covered, polish missing
2 Weak, incomplete, inconsistent
1 Missing or actively harmful

Чек-листы по подсистемам

1. Instructions (1-5)

  • CLAUDE.md существует
  • CLAUDE.md ≤200 строк (если больше — балл -1)
  • domain-rules.yaml существует и заполнен (не template)
  • domain-rules.yaml имеет invariants ≥2
  • domain-rules.yaml.freshness.last_reviewed более 30 сессий назад
  • docs/PRODUCT.md существует
  • docs/ARCHITECTURE.md существует

2. State (1-5)

  • feature_list.json валиден JSON
  • feature_list.json имеет active либо явно null
  • WIP=1: только одна active (если есть)
  • feature_list.json не пустой (есть features in some state)
  • SESSION.md обновлён более 7 дней (внешний календарь)
  • SESSION.md имеет current state filled
  • Нет дубля state между файлами (.planning/STATE.md, CHECKPOINT.md и т.п. — старые v3/v4 артефакты которых не должно быть в v5.1)

3. Verification (1-5)

  • feature.verification объявлен для done фичей (4-layer)
  • Verification commands явно прописаны (не TODO)
  • eval-samples/ существует с реальными сценариями
  • Negative-verification doced (verification_self_check date)
  • User-reported layer защита упоминается где-то
  • Test-strategy.md есть для активной фичи

Read the full file on GitHub · 167 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 · 167 lines · 69 tokens per session scan A b9f68b56c7e3

Subscribe to this mod's changes

evaluator-agent is an agent published in the GitHub repository andrewcigan/vibe-dev-plugin (5 stars, last pushed 1mo ago), licensed MIT. It adds 69 tokens to every session and 1,829 once invoked, about $0.0003 per session on Opus 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.