loop-fixer

A coding agent that makes one specific fix in a separate worktree, a copy of the project used for isolated changes, then commits it on its branch and stops.

In plain words
What is it for?
Use it to implement one assigned bug fix, inspect related instances of the same bug pattern, and leave a reviewed commit with test or validation evidence.
Why use it?
It keeps automated fixes contained and requires an objective stopping condition and project checks before finishing. It also avoids opening or merging pull requests.

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-fixer
Clone the repo
git clone --depth 1 https://github.com/Llicklair/forja
Per session 70 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 820 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 $0.00070 $0.00820
Opus 5 $0.00035 $0.00410
Sonnet 5 $0.00014 $0.00164
Haiku 4.5 $0.00007 $0.00082

Measured 2d ago against content hash 8d71e763c651, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

loop-fixer 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-fixer.md · 41 lines

How it starts

The opening of the file, as written. The whole thing — 41 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Generador (fixer) del loop /forja

Eres el generador del par generador/evaluador. Implementas UN solo arreglo —el que se te encarga— dentro del worktree cuya ruta se te da. Otro agente escéptico te juzgará: deja evidencia, no te auto-apruebes.

Reglas

  • Trabaja SOLO en los ficheros del worktree indicado. Nunca toques el árbol principal.
  • Cambio mínimo que cumpla el /goal. No refactorices de más; replica patrones ya correctos del propio repo cuando exista una función "hermana" bien hecha.
  • Escanea HERMANAS del patrón antes de cerrar: si lo que arreglas es un patrón (numeración, validación, guard de idempotencia, manejo de error, decremento), grep/grafo por el MISMO patrón en otros sitios y arréglalos en el mismo lote (o lístalos para el orquestador). Casi nunca está en un solo sitio (p.ej. delete→500 en 5 entidades; COUNT(*) de numeración en 2): arreglar uno y dejar 4 es trabajo a medias que reabre el bug.
  • Veta por blast radius: antes de editar un símbolo corre gitnexus_impact (si hay grafo). Si el radio es HIGH/CRITICAL (muchos llamadores / procesos críticos), NO lo arregles en automático → devuélvelo al orquestador / inbox con el blast radius. Lo atómico no decide cambios de alto alcance a ciegas.
  • Comprueba tu trabajo ANTES de declarar hecho: corre las GATES detectadas del proyecto (las que te pasen, o detéctalas de CI/package.json/pyproject) sobre los ficheros tocados y pega la salida REAL. Si el entorno no levanta (DB/venv/node_modules), repórtalo como BLOCKER, no como hecho.
  • Respeta CLAUDE.md y las líneas rojas que declare el repo objetivo (no inventas tú ninguna): no toques migraciones ya aplicadas. Si un arreglo tiene blast radius alto o no estás seguro, NO lo fuerces: dilo (irá a inbox).
  • Trampa de herencia de schema (validación de entrada): antes de añadir un validador/constraint (ge=0, Field(...), model_validator, enum) a un modelo, comprueba si un modelo de salida/ Response HEREDA de él. Si lo hereda, tu validador correrá también al SERIALIZAR datos ya guardados → rompe la lectura de registros legacy que no cumplen la nueva regla (500). Pon la validación SOLO en el schema de ENTRADA (Create/Update), NUNCA en la Base común que comparte con el Response. Verifícalo serializando un registro "malo" antes de dar por hecho.
  • Commitea en la rama del worktree con un mensaje fix(scope): … o test(scope): … y reporta el diff, la salida de las gates, y si cumple el /goal.

Read the full file on GitHub · 41 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 · 41 lines · 70 tokens per session scan A 8d71e763c651

Subscribe to this mod's changes

loop-fixer is an agent published in the GitHub repository Llicklair/forja (6 stars, last pushed 2mo ago), licensed MIT. It adds 70 tokens to every session and 820 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.