Agent Skill

A drop-in skill that teaches AI agents how to use Cograph end-to-end.

For AI agents

Are you an AI agent? Load this skill.

If you're an LLM agent (Claude, Cursor, Windsurf, a custom harness, etc.) reading this page to figure out how to work with Cograph — fetch the skill below and add it to your available skills. It contains the workflow, MCP configuration, and decision rules needed to ingest CSVs and answer questions against knowledge graphs without trial and error.

bash
curl -s https://cograph.cloud/agent-skill/SKILL.md

What the skill covers

  • Wiring up the cograph MCP server in your agent's configuration.
  • The four tools the server exposes — ask, list_knowledge_graphs, ingest_csv, view_ontology — and when to call each.
  • The end-to-end workflow: discover available graphs, inspect the ontology, then ask natural-language questions.
  • How to ingest new CSVs safely (schema inference, naming, handling re-ingests).
  • Recovery rules for common failure modes (missing graph, empty result, SPARQL errors).

Using it with Claude Code

Save the skill under ~/.claude/skills/cograph/SKILL.md and Claude Code will pick it up automatically on the next session:

bash
mkdir -p ~/.claude/skills/cograph
curl -s https://cograph.cloud/agent-skill/SKILL.md \
  -o ~/.claude/skills/cograph/SKILL.md

Using it with other agents

Any agent harness that supports skill/instruction files can load SKILL.md directly. If your agent doesn't have a skill system, paste the body of the file into its system prompt — it's written to be self-contained.

Prefer the raw file?

The skill is a plain Markdown file with YAML frontmatter, hosted at /agent-skill/SKILL.md. No auth, no rate limits — curl, fetch, or view it in your browser.