El sistema operativo de conocimiento para agentes de IA
Conocimiento que estructura.
Agentes que entienden.
Software que mejora.
AKOS organiza el conocimiento técnico y lo convierte en procesos reproducibles que los agentes pueden ejecutar, verificar y mejorar continuamente.
git clone https://github.com/ipaud/AKOS.git ~/DEV/AKOS
cd ~/DEV/AKOS
./install.sh
./doctor.shCifras del repositorio auditado. No incluyen métricas de adopción ni comerciales.
Propuesta de valor
Tus agentes ya saben programar. AKOS les enseña cómo trabajar.
Menos improvisación. Más criterio operativo. El contexto explica el proyecto; AKOS define cómo trabajar sobre él.
Conocimiento gobernado
13 dominios, packs versionados, revisiones trazables.
Procesos reproducibles
Workflows y reglas que no dependen del humor del prompt.
Validación real
Benchmarks, tests y checklists en cada revisión.
Flujo operativo
Seis pasos, un solo criterio
Cada sesión de AKOS sigue el mismo ciclo determinista. Nada de resultados sorpresa.
- 01
Detecta contexto
Reconoce lenguajes, frameworks y patrones del repositorio.
- 02
Selecciona conocimiento
Activa los Knowledge Packs relevantes al problema.
- 03
Activa skills y agentes
Combina revisores especializados según el workflow.
- 04
Analiza
Ejecuta checklists, reglas y comprobaciones deterministas.
- 05
Valida
Contrasta hallazgos con benchmarks y evaluaciones.
- 06
Registra
Escribe historial trazable para auditoría y mejora continua.
Knowledge Packs
Conocimiento operativo, no genérico
Cada pack define principios, patrones, riesgos, checklists y fuentes. Todos versionados.
Personal Layer — Pau Avila
Capa personal de autoridad Level 0: convenciones, preferencias y patrones propios que todo agente cargado con AKOS aplica automáticamente. No es la destilación de una fuente externa; solo la anula el safety floor (seguridad, accesibilidad básica, integridad de datos) y los requisitos legales.
Agent Evals — Prove It Got Better
Operationalizes the measurement of an agent: whether a change made it better, and whether anyone would find out if it made it worse. The governing idea: do not accept an agent improvement backed by three manual examples — demand a representative suite and a comparison against a stated baseline. This is the discipline traditional testing already applies to deterministic code, applied to non-deterministic, judgment-laden output. Structurally almost everything transfers — fixed inputs, a runner, CI, red-before-green, coverage as a question worth asking — and exactly one thing does not: the assertion. A deterministic system lets you compare exact values; a judgment-laden one needs a grader that tolerates variation and still says something falsifiable. Everything hard about evals lives in that substitution, and everything else is a solved problem being re-solved badly.
Agent Foundations — Which Shape, and How Bounded
Operationalizes the choice of architecture for a system built on a language model, and the discipline of keeping the chosen one under control. The governing idea: don't reach for an agent when a deterministic function, a fixed workflow, or a direct query solves the problem. Between one model call and a full agent lies a spectrum of shapes — chains, routers, fan-outs, workflows, orchestrators, critique loops — and each step along it transfers a decision from the engineer to the model, buying adaptability with predictability, testability, and a cost ceiling. Most of the work is knowing which shape fits the task. The rest is making a warranted agent terminate on a stated condition, stay inside budgets that fail loudly, recover by the right response to the right class of failure, and take only actions that are safe to take twice.
Agent Security — Untrusted Text, Bounded Actions
Operationalizes the security of a system where a language model plans, calls tools, and acts. The governing idea: retrieved text, documents, web content, and tool output are untrusted data, never instructions — a boundary application security has enforced for thirty years at every other layer, and which language models collapse by construction, because instructions and data are the same substrate read by the same mechanism. From that collapse follow prompt injection in both forms, tool and resource poisoning, exfiltration through channels nobody classified as egress, memory that stays poisoned across sessions, and a confused deputy acting with permissions its requester never held. The remedy is not a better prompt. It is bounding what the agent CAN do — scoped credentials, sandboxes, egress allowlists, approval gates on irreversible actions, and an audit record — so that a model persuaded by a hostile document achieves nothing outside those bounds.
Agentes especializados
Una revisión completa necesita más de una perspectiva
Cada agente aporta un ángulo distinto: seguridad, arquitectura, UX, accesibilidad, rendimiento, datos, IA.
Audits UI against WCAG 2.2 AA — the safety-floor accessibility check no profile or preference can waive. Verifies keyboard operability, accessible names, contrast, focus, semantics, and error handling.
Reviews structural decisions: right-sized layering, dependency direction, coupling/cohesion, and — critically — challenges unnecessary complexity. Guards both under-engineering (missing boundaries where volatility is real) and over-engineering (ceremony without payoff).
Reviews backend/API code: REST/GraphQL design correctness, resolver/query performance, and the twelve-factor deployment discipline. Security-specific concerns route to security-reviewer; DB-specific to database-reviewer.
Reviews interface copy: is every word earning its place? Are labels honest, specific, and obvious? Cuts happy talk, vague labels, and unread instructions.
AKOS en acción
Terminal y reporte, sincronizados
Ejecuta un workflow y obtén un informe accionable con severidad, contexto y trazabilidad.
- Comandos deterministas por workflow.
- Salida legible y estructurada.
- Integra con CI y con tu editor.
Sin AKOS · Con AKOS
Del prompt al proceso
La diferencia entre respuestas puntuales y un sistema que aprende.
- · Cada sesión reinventa criterios.
- · Los agentes “creen que saben” pero no justifican.
- · Los hallazgos no dejan rastro auditable.
- · Revisiones dependen del prompt del día.
- · Deuda invisible hasta que estalla en producción.
- · Packs versionados definen el criterio.
- · Los agentes citan la regla que aplican.
- · Historial estructurado por sesión.
- · Workflows deterministas, no improvisados.
- · Benchmarks continuos detectan regresiones.
Integraciones
Tool-agnostic. Markdown-first.
AKOS no sustituye tus herramientas: las potencia con conocimiento.
Construye agentes con criterio, no solo con contexto.
Conocimiento estructurado. Decisiones trazables. Resultados verificables.