All Docs
Claude Code Skill
The Cognova skill is the lightest way to connect Claude Code to your workspace. It's a Python script that calls the REST API and outputs Markdown directly into Claude's context window — no protocol overhead.
Install
Run this in any project where you want Cognova access:
mkdir -p .claude/skills/cognova
curl -o .claude/skills/cognova/fetch.py https://cognova.dev/api/v1/skill
curl -o .claude/skills/cognova/SKILL.md https://cognova.dev/skill/SKILL.md
python3 .claude/skills/cognova/fetch.py --setup
The --setup command prompts for your API URL and API key, then saves a config.json in the skill directory.
Commands
Agents
# List all workspace agents
python3 .claude/skills/cognova/fetch.py agents
# Ask the default agent
python3 .claude/skills/cognova/fetch.py ask "What authentication decisions have we made?"
# Ask a specific agent by name
python3 .claude/skills/cognova/fetch.py ask code-reviewer "Review this function"
Tasks
# List active tasks
python3 .claude/skills/cognova/fetch.py tasks
# List by status
python3 .claude/skills/cognova/fetch.py tasks list in_progress
# Create a task
python3 .claude/skills/cognova/fetch.py tasks create "Implement user search"
# Complete a task
python3 .claude/skills/cognova/fetch.py tasks complete TASK_ID
Memory
# Search memories
python3 .claude/skills/cognova/fetch.py recall "authentication decisions"
# Store a memory
python3 .claude/skills/cognova/fetch.py remember "We chose JWT over sessions for the API"
Knowledge
# List knowledge files
python3 .claude/skills/cognova/fetch.py knowledge
# Read a specific file
python3 .claude/skills/cognova/fetch.py knowledge docs/architecture.md
Diagnostics
# Check connection status
python3 .claude/skills/cognova/fetch.py --status
# Show help
python3 .claude/skills/cognova/fetch.py --help
How It Works
The skill is a self-contained Python script that:
- Reads
config.jsonfrom the skill directory for the API URL and key - Uses
curl(viasubprocess) to call the Cognova REST API - Formats responses as Markdown and prints to stdout
- Claude Code reads stdout directly into its context
No dependencies beyond Python 3 and curl.
Skill vs MCP
| Feature | Skill | MCP |
|---|---|---|
| Setup | 3 commands | 1 command |
| Protocol | HTTP + stdout | MCP Streamable HTTP |
| Tool discovery | Manual commands | Automatic |
| Context overhead | Minimal | Higher |
| Best for | Quick lookups, task management | Full bidirectional integration |
Use the skill when you want lightweight, focused access. Use MCP when you need full tool discovery and richer integration.