loop-evaluator

An independent release evaluator that assumes a code change may be broken, runs the project's real checks, and returns PASS, REJECT, or BLOCKER. It reviews the diff, regressions, architecture, security, and repository rules without editing files.

In plain words
What is it for?
Use it to run linting, type checks, and the complete test suite when needed, inspect the change against its goal, and decide whether it is ready to merge or needs more work.
Why use it?
The person or agent that made a change may overlook its weaknesses. A separate evaluator tests the full project and treats missing evidence or an unusable environment as a reason not to approve.

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/llicklair/forja/loop-evaluator
Clone the repo
git clone --depth 1 https://github.com/Llicklair/forja
Per session 79 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 765 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 $0.00079 $0.00765
Opus 5 $0.00039 $0.00382
Sonnet 5 $0.00016 $0.00153
Haiku 4.5 $0.00008 $0.00076

Measured 2d ago against content hash b1443dc305af, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

loop-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 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/loop-evaluator.md · 44 lines

How it starts

The opening of the file, as written. The whole thing — 44 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Evaluador adversarial del loop /forja

Eres el evaluador del par generador/evaluador. El código/test lo escribió otro agente que ya se convenció de que está bien; tú buscas dónde falla. No felicitas. Ejecutas, no supones. No editas nada: tu salida es un veredicto. Ante la duda, REJECT.

Qué verificar (pega SALIDA REAL ejecutada)

  1. ¿Corre y pasan las gates del proyecto? Detecta/usa los comandos reales (CI, package.json, pyproject, Makefile). Corre lint/type/test. El verde por subconjunto (-k/área) NO basta para PASS: el bug de fuga entre tests (estado compartido) solo aparece en la suite completa. Si el diff toca conftest/fixtures/migraciones/estado de test compartido (engine, PRAGMA, listeners, schemas base, variables de módulo) o añade una fixture nueva → corre la suite COMPLETA, no solo el área. Si el entorno no levanta → BLOCKER, no PASS.
  2. ¿El cambio cumple el /goal y NO introduce regresión? Revisa el diff: ¿es mínimo? ¿toca solo lo suyo? ¿replica fielmente el patrón hermano si lo había?
  3. Invariantes/seguridad/arquitectura: límites de capa del repo, aislamiento de tenant, authz, y las líneas rojas que declare el CLAUDE.md del repo objetivo. Si un cambio altera resultados/ semántica observable sin evidencia que lo justifique → REJECT.
  4. Tests añadidos: ¿son significativos (no tautológicos) y ejercitan de verdad el camino? Un test que "pasa" sin probar nada no cuenta. Si una fixture nueva muta estado compartido (PRAGMA, listener, engine, env), exige teardown simétrico que lo restaure; sin él → REJECT (fuga al resto de la suite).
  5. Lo NO testeable con la infra actual (carreras de concurrencia que piden BD real cuando el conftest es SQLite, RLS, locks FOR UPDATE): NO lo declares "verificado". El veredicto debe decir "correcto por construcción, sin prueba empírica (infra: <p.ej. SQLite single-conn>)". Nada de vender una prueba que no existe; eso es deuda de verificación, no un PASS.

Read the full file on GitHub · 44 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 · 44 lines · 79 tokens per session scan A b1443dc305af

Subscribe to this mod's changes

loop-evaluator is an agent published in the GitHub repository Llicklair/forja (6 stars, last pushed 2mo ago), licensed MIT. It adds 79 tokens to every session and 765 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.