trio-evaluator

trio-evaluator is an agent for Claude Code from albiol2004/trio-agent-loop. It costs 85 tokens per session (3,615 once invoked), scanned A, original, MIT.

An adversarial evaluator for a two-agent coding loop in which a Lead builds an iteration and an Evaluator checks it. It independently tests the code against the acceptance criteria and writes a SHIP or ITERATE verdict in VERDICT.md.

In plain words
What is it for?
Use it to verify a coding iteration against PLAN.md, run builds and tests, inspect the Git diff, and identify unmet criteria. It reports problems for another iteration instead of changing the code itself.
Why use it?
It reduces the risk of accepting a confident but incorrect implementation. The evaluator checks the actual working tree and test results before reviewing the Lead's report.

Agent for Claude Code

Written for Claude Code: effort in frontmatter. Also seen: model in frontmatter; mentions subagents.

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/albiol2004/trio-agent-loop/trio-evaluator
Clone the repo
git clone --depth 1 https://github.com/albiol2004/trio-agent-loop

Made for: Claude Code.

Wrote 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.

agentmods badge for trio-evaluator

README.md
[![agentmods](https://agentmods.dev/badge/agents/albiol2004/trio-agent-loop/trio-evaluator.svg)](https://agentmods.dev/agents/albiol2004/trio-agent-loop/trio-evaluator)
Your own site
<a href="https://agentmods.dev/agents/albiol2004/trio-agent-loop/trio-evaluator"><img src="https://agentmods.dev/badge/agents/albiol2004/trio-agent-loop/trio-evaluator.svg" alt="Measured on agentmods" height="20"></a>
Per session 85 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,615 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 5.1 $0.00085 $0.03615
Opus 5 $0.00043 $0.01808
Sonnet 5 $0.00017 $0.00723
Haiku 4.5 $0.00009 $0.00362

Measured today against content hash 86b4120ec557, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

trio-evaluator 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 today.

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.

.claude/agents/trio-evaluator.md · 219 lines

How it starts

The opening of the file, as written. The whole thing — 219 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Role: Evaluator (adversarial verify) — one iteration

You are the Evaluator in a two-agent loop (Lead → Evaluator), equal in rank to the Lead. You are adversarial by design: your job is to find the ways the iteration is wrong, not to confirm it is right. You never fix code — a broken build gets an ITERATE verdict, not a patch.

The orchestrator's prompt may name a mailbox directory other than loop/ (and/or a project root other than your cwd) — if it does, resolve every loop/ path below there. Never touch any other loop* directory you find in the tree: it belongs to a different loop.

Inputs — ORDER MATTERS (anti-sycophancy protocol)

Form your own verdict BEFORE reading the Lead's claims. Same-model judges over-trust a confident report; don't give it the chance.

  1. loop/GOAL.md — the mission (immutable; overrides everything else).
  2. loop/PLAN.md — the acceptance criteria are your checklist. Check them verbatim.
  3. The working tree — the actual diff (git diff, git status) and your own execution of builds/tests.
  4. Only after you have per-criterion results: read loop/REPORT.md and check it for discrepancies against what you observed. A claim you did not reproduce stays unverified.

Context gathering — evaluate from knowledge, not vibes

Build real context before judging; fan out Sonnet trio-scout subagents in parallel via the Agent tool. The Evaluator itself remains Opus; all scoped exploration and mechanical support remains Sonnet:

  • Blast radius: call sites of changed functions, conventions the diff violates, dead code left behind, side effects elsewhere in the repo.
  • API currency: for each significant library/API the diff touches, check (via WebSearch/WebFetch or scouts) that the code uses the current recommended API for the version actually pinned in this project — not a deprecated pattern from stale training data. Flag deprecated/removed APIs, known CVEs in newly added dependencies, and version mismatches between what the code assumes and what the lockfile/manifest pins. Judge against the project's pinned versions, not the newest thing on the internet — "not the latest major" alone is a non-blocking observation, "deprecated in the pinned version" is blocking.

Read the full file on GitHub · 219 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. today Changed · +39 lines 86b4120ec557
  2. 5d ago First seen · 180 lines · 85 tokens per session scan A b9a8c4ed7f34

Subscribe to this mod's changes

trio-evaluator is an agent published in the GitHub repository albiol2004/trio-agent-loop (2 stars, last pushed 5d ago), licensed MIT. It adds 85 tokens to every session and 3,615 once invoked, about $0.0004 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.

Related

Other agents, from other repositories

blind-acceptance

Слепая приёмка перед словом «готово». Получает ТОЛЬКО исходную просьбу человека и сделанные изменения — без спеки, плана, тикетов и разговора — и отвечает на единственный вопрос: сделали ли то, о чём просили. Вызывать в конце работы, после того как гейт верификации дал ноль. Судит соответствие, а не качество; ничего…

edvardgrishin27/superstack-plugin · 0 tokens

builder

Пишет код одного таска в свежем контексте и возвращает контракт. Получает готовый промпт от handoff.py — ticket, границы уже построенного, разделы спецификации, тестовый контракт. Вызывать на каждый таск отдельно; двух тасков в одном контексте не бывает. Не оркестрирует, не правит чужие зоны, не решает за человека.

edvardgrishin27/superstack-plugin · 87 tokens

second-opinion

Читающий-только советчик на самой сильной модели. Вызывать на границах решения — до того, как выбрана архитектура, схема данных, форма API или стратегия рефакторинга; когда одна и та же задача не поддалась двум разным попыткам; и ОБЯЗАТЕЛЬНО один раз перед тем, как сказать «готово». Получает решение (или диф) и…

edvardgrishin27/superstack-plugin · 0 tokens

cpp-reviewer

Expert C++ code reviewer specializing in memory safety, modern C++ idioms, concurrency, and performance. Use for all C++ code changes. MUST BE USED for C++ projects.

affaan-m/ECC · 41 tokens

dynamic-agents

Dynamic agents use functions instead of static values for instructions, model, and tools. These functions receive runtime context and return the appropriate configuration for each operation.

VoltAgent/voltagent · 0 tokens

design-rules

Condensed 10 Golden Rules from the Agent Design Bible.

jmagly/aiwg · 0 tokens