Arya Model Router

Efficiently routes tasks between cheap, default, and pro models using optional sub-agents and briefing to minimize token use and cost.

Security Scan

The skill's code, instructions, and data are coherent with a local model-routing helper: it reads/writes only its own files, makes routing decisions locally, and does not request credentials or call external services itself.

Purpose & Capability

The files and runtime instructions match the stated purpose: router.py implements local routing/thresholds/overrides and brief.py creates a compact brief. The rule set lists models and thresholds consistent with a model router. Minor inconsistency: SKILL.md metadata lists required bins (bash, python3) while the registry 'Requirements' section reported none — this is a small metadata mismatch (python3 is logically required to run the scripts).

Instruction Scope

SKILL.md and README instruct running local scripts (router.py, brief.py) and interpreting JSON output. The scripts only read/edit local files (rules.json, state.json) and do no network or model calls themselves. Important operational note: brief.py extracts lines that look like URLs, job IDs, stack traces, etc.; if the downstream integration sends the brief to a remote model (e.g., an OpenAI model recommended by the router), that brief could contain sensitive tokens/IDs/URLs you supplied in the context. The router itself does not exfiltrate data, but the decisions it emits may cause other components to send data to external models.

Install Mechanism

No install spec (instruction-only skill with bundled scripts). No downloads or external installers are used; files are local and static, which is lower risk.

Credentials

The skill requests no environment variables or credentials and does not access system-wide config. The SKILL.md metadata's 'bins' requirement (bash, python3) is reasonable for running the included scripts. There are no unrelated secrets requested.

Persistence & Privilege

The skill writes and updates a local state file (state.json) in its own directory to store mode/feedback/lastDecision — this is scoped to the skill and expected. always is false and the skill does not modify other skills or system-wide agent settings.

Assessment

This skill appears to be what it says: a local router that decides when to escalate to stronger models. Before installing, consider: 1) It writes rules.json and state.json in the skill folder — review and back up those files if you edit them. 2) brief.py will surface URLs, job IDs, errors and similar 'interesting' lines; do not pass sensitive tokens/credentials in context you expect to be briefed and potentially forwarded to an external model. 3) The router only emits decisions (model names like openai/gpt-4.1); actual calls to remote LLMs happen elsewhere — ensure any downstream integration that executes 'use_subagent' is authorized and has appropriate API keys and data-handling policies. 4) Ensure python3 is available (SKILL.md lists bash and python3). 5) Optionally review and customize rules.json thresholds/keywords to match your workload and to avoid unnecessary escalation. If you want a second check, provide how your agent invokes sub-agents (what component accepts 'use_subagent' and whether it will send briefs to remote models).

Arya Model Router (Token Saver)

Router de modelos para OpenClaw: decide cuándo usar un modelo barato vs uno más fuerte, reduciendo costo y tokens.

Objetivos

• Mantener el chat diario barato.
• Escalar a un modelo superior solo cuando la tarea lo amerite.
• Evitar pasar contexto enorme al modelo caro: primero crear un brief.

Enfoque

• El agente principal (main) se mantiene en un modelo económico.
• Para tareas pesadas, el router recomienda (o ejecuta) sub-agentes con un modelo superior.

Niveles (por defecto)

• cheap: openai/gpt-4o-mini
• default: openai/gpt-4.1-mini
• pro: openai/gpt-4.1

Uso (conceptual)

• "Router: responde esto en modo cheap" (forzado)
• "Router: analiza esto" (auto)

Archivos

• router.py: clasificador + reglas
• rules.json: reglas editables
• README.md: documentación completa