Skip to content
This repository was archived by the owner on Mar 23, 2026. It is now read-only.

AGENTS.md

Latest commit

 

History

History
139 lines (116 loc) · 6.37 KB

AGENTS.md

File metadata and controls

139 lines (116 loc) · 6.37 KB

TECHKNOWMAD LABS — Cortex Skill Library

27 self-evolving AI research skills built on Karpathy's autoresearch pattern. Each skill has an ARENA.md that defines its evolution strategy — skills compete in an evaluation arena, mutate based on LLM-as-Judge scores, and the best variants survive. The Skill Organism engine runs genetic selection across the full ecosystem. Includes trilogy integration (MindSpider social listening, intelligence analysis, multimodal analysis, forum intelligence, scenario simulation) and cross-platform adapters. Every skill runs as a Claude Skill, MCP Server, LangChain Tool, CrewAI Tool, or OpenAI GPT Action from a single source definition.

Skills

Code Quality & Security

  • security-audit: SAST scanning (bandit + semgrep) with secret detection and severity-graded reports
  • code-review-engine: Multi-dimensional code review (correctness, security, performance, maintainability, style)
  • de-slop: AI writing pattern detector — scores 0-100 for emoji abuse, hyperbolic language, buzzword stacking
  • tdd-enforcer: TDD compliance enforcement with coverage thresholds, anti-pattern detection, and pre-commit hooks

Agent Infrastructure

  • agent-orchestrator: Multi-agent workflow orchestration with dependency resolution and parallel execution
  • agent-output-validator: Schema/quality/safety validation for agent outputs — catches hallucinations and drift
  • context-engineer: Optimal context window assembly with compression, prioritization, and token budgeting
  • persistent-memory: Key-value memory with TTL, tagging, and semantic search across sessions
  • session-memory: Lightweight session-scoped checkpoints for context window resets

Development Lifecycle

  • dev-lifecycle-engine: Full SDLC tracking from planning through production with gate checks
  • diff-generator: Snapshot-based file system diffs with structured change reports
  • github-mcp: GitHub API integration (repos, files, branches, PRs, issues, releases) via MCP
  • repo-publisher: Pre-publish pipeline chaining security, slop detection, structure validation, and metadata updates

Research & MLOps

  • research-workflow: Experiment design, methodology enforcement, and structured research output
  • mlops-standards: ML operations best practices enforcement with reproducibility patterns
  • prompt-architect: Prompt engineering and optimization with structured prompt templates

Intelligence & Data (Trilogy Integration)

  • mindspider-connector: Live social listening feed connector with demo/MySQL modes
  • intelligence-query: Multi-source intelligence analysis with topic correlation and trend detection
  • multimodal-analyst: Cross-modal content analysis (text, image, video) with hallucination warnings
  • forum-intelligence: Forum thread analysis with coordination detection and minority viewpoint extraction
  • scenario-simulator: MiroFish-inspired swarm scenario simulation with counterfactual injection

Skill Development

  • meta-skill-evolver: Automated skill analysis, metrics collection, and iterative self-improvement
  • skill-validator: Structure and frontmatter validation against canonical skill spec
  • skill-test-harness: Test framework for skills with coverage reporting and 8 assertion types
  • pre-package-pipeline: Pre-packaging validation ensuring .skill file readiness
  • design-system-forge: Design system generation from brand parameters with accessibility compliance

Quick Start

Claude Code (Native)

cd skills/<skill-name>
# Skills load automatically via SKILL.md

MCP Server (Claude, Copilot, Cursor, Windsurf, VS Code, JetBrains)

cd cross-platform/generated/mcp/<skill-name>
uv run python server.py

LangChain / LangGraph

from langchain.<skill_name>.tools import get_all_tools
tools = get_all_tools()
agent = create_react_agent(llm, tools)

CrewAI

from crewai.<skill_name>.tools import *
agent = Agent(role="...", tools=[SecurityScanTool()])

OpenAI Custom GPT

Import openai/<skill-name>/openapi.json as a GPT Action in the GPT Builder.

Architecture

cortex-research-suite/
├── skills/                     # Source skills (SKILL.md + scripts/ + references/)
├── cortex/                     # Python framework (evaluation, agents, synthetic data)
├── cross-platform/
│   ├── manifests/              # Universal Skill Manifests (JSON)
│   ├── adapters/               # Platform adapter generators
│   └── generated/              # Platform-specific outputs
│       ├── mcp/                # FastMCP servers
│       ├── langchain/          # LangChain BaseTool classes
│       ├── crewai/             # CrewAI BaseTool classes
│       ├── openai/             # OpenAPI 3.1 specs for GPT Actions
│       └── agents/             # AGENTS.md discovery files
├── skill-organism/             # Skill evolution engine
├── packages/                   # .skill files for Claude Desktop
└── AGENTS.md                   # This file

Universal Skill Manifest

Each skill is defined by a single JSON manifest that maps to all platforms:

{
  "name": "security-audit",
  "version": "1.0.0",
  "capabilities": [
    {
      "name": "security_scan",
      "input_schema": { "type": "object", "properties": {} },
      "output_schema": { "type": "object", "properties": {} }
    }
  ],
  "platforms": {
    "claude_skill": { "entry_point": "SKILL.md" },
    "mcp_server": { "command": "uv", "args": ["run", "python", "server.py"] },
    "langchain": { "tool_class": "SecurityScanTool" },
    "crewai": { "tool_class": "SecurityScanTool" },
    "openai_gpt": { "openapi_spec": "openai/security-audit/openapi.json" }
  }
}

Supported Platforms

Platform Transport Status
Claude Desktop / Claude Code Skill (.skill) Production
Claude Desktop MCP (stdio) Production
GitHub Copilot MCP (stdio) Production
Cursor MCP (stdio) Production
Windsurf MCP (stdio) Production
VS Code (Copilot) MCP (stdio) Production
JetBrains IDEs MCP (stdio) Production
LangChain / LangGraph Python SDK Production
CrewAI Python SDK Production
OpenAI GPTs OpenAPI Action Production
Google A2A Agent Card Available

License

MIT — TECHKNOWMAD LABS https://github.com/TECHKNOWMAD-LABS