convez.ai
MCP ServerAutomaçãoIntermediário
cognis-digital

Cognis Skills — Registro de Capacidades para Agentes IA

cognis-digital/skills

Registry de skills model-agnostic para agentes IA descobrirem e executarem capacidades

Ver no GitHub
BomPluginScore · de 0 a 100

Registro portátil de skills (capacidades) que agentes IA podem descobrir, carregar e invocar em runtime. Cada skill é um diretório autocontido com manifesto YAML, script executável e documentação, seguindo o padrão ClawHub/Claude-skills. Facilita orquestração de tarefas como busca web, auditoria de repositórios, sumarização e análise de segurança.

Instalação

$npx convez-ai cognis-digital/skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
100%
Último commit
há 4 meses
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Orquestração de múltiplas capacidades em agentes autônomos (ATD trader, fleets)
  • Criação de skill marketplace portável entre diferentes plataformas de IA
  • Auditoria e análise automática de repositórios, segredos e SQL
  • Busca e pesquisa com fallback keyless via DuckDuckGo/RSS

Veredito

Solução bem arquitetada para federação de skills em agentes. Manifesto YAML + contrato JSON claro reduz fricção. Útil para times que precisam manter catálogos de capacidades reutilizáveis; limitação é depender de definição manual de entrypoints e permissões.

README

A tradução para o português deste README ainda está na fila.

cognis-skills

An agent skill registry for Cognis Digital LLC autonomous agents (ATD trader, cog4 fleet, Mission Control). Skills are self-contained, model-agnostic capabilities an agent can discover, load, and invoke at runtime — in the spirit of ClawHub / Claude-skills manifests.

What a skill is

A skill is a directory under skills/ containing:

FileRequiredPurpose
SKILL.mdyesHuman + machine-readable manifest with YAML frontmatter + usage docs
a script (run.py, run.sh, ...)yesThe executable that performs the work
supporting filesnofixtures, schemas, prompt templates

The top-level registry.json indexes every skill so an agent can resolve a name → directory → entrypoint without scanning the tree.

SKILL.md format

Each manifest opens with YAML frontmatter, then free-form Markdown documentation:

---
name: web-search
version: 1.0.0
description: One-sentence trigger so the planner knows WHEN to use this skill.
entrypoint: run.py
runtime: python3
args:
  - name: query
    type: string
    required: true
    description: Search query string.
inputs: { stdin: false }
outputs: { format: json }
permissions: [network]
tags: [research, osint]
---

Frontmatter fields

  • name — unique, kebab-case. Matches the directory name and the registry.json key.
  • version — semver.
  • description — a trigger sentence. Written so a planner LLM can decide when to invoke without reading the body.
  • entrypoint — script filename, relative to the skill directory.
  • runtime — python3, bash, etc. Determines how the loader execs the entrypoint.
  • args — ordered list; each has name, type, required, description. Passed as --name value CLI flags.
  • permissions — capability tags the host must grant: network, filesystem, read-only, subprocess.
  • tags — free-form discovery labels.

Invoking a skill

Resolve the entrypoint from the registry and exec it. Scripts are stdlib-only and emit JSON on stdout:

python3 skills/web-search/run.py --query "critical minerals export controls"
python3 skills/repo-audit/run.py --path .
python3 skills/secret-scan/run.py --path ./src

Programmatic loading:

import json, subprocess, sys
from pathlib import Path

reg = json.loads(Path("registry.json").read_text())
skill = reg["skills"]["secret-scan"]
entry = Path("skills") / skill["name"] / skill["entrypoint"]
out = subprocess.run([sys.executable, str(entry), "--path", "."],
                     capture_output=True, text=True)
result = json.loads(out.stdout)

Contract

Every skill MUST:

  1. Accept its declared args as --name value flags.
  2. Print a single JSON object to stdout (machine output).
  3. Print diagnostics to stderr only.
  4. Exit 0 on success, non-zero on hard failure.
  5. Run on stdlib alone unless a requirements field declares otherwise (none here do).

Skills in this registry

SkillDescription
web-searchKeyless web/news search via DuckDuckGo + RSS, returns ranked results.
repo-auditStatic health audit of a repo (size, languages, large files, missing meta).
summarizeExtractive summarizer for text/Markdown files.
sql-explainParse and explain a SQL statement in plain English.
secret-scanRegex + entropy scan for leaked credentials.
changelogGenerate a changelog section from git history.
compliance-checkCheck a repo for required policy/license/security files.
osint-lookupResolve a domain/host to public footprint signals.

How it fits

flowchart LR
  U[You / CI / Agent] --> R[skills]
  R --> O[Outputs & artifacts]
  R --> M[MCP / JSON]
  M --> AI[AI agents]
  R --> S[Cognis Neural Suite]

Explore the suite → 🗂️ all tools · ⭐ awesome-cognis · 🔗 cognis-sources

License

MIT. (c) 2026 Cognis Digital LLC.

Ver o original no GitHub

Bom para

Relacionados