gocam-agent

An AI-assisted workspace for exploring, reviewing, and editing GO-CAM models. Open this repo in Claude Code and start working with GO-CAM models through natural language.

⚠️ This workspace targets the development server by default. Production models require the --live flag.

Getting started

git clone https://github.com/geneontology/gocam-agent.git
cd gocam-agent
uv sync
export BARISTA_TOKEN=<your-token>
claude

Then just ask:

• "What GO-CAM models exist for human TP53?"
• "Show me the most recent models and summarize them"
• "Review model 680ad14200006567 — is it well-structured?"
• "Help me add causal edges to connect the enzymatic steps in this sphingolipid model"

The agent has access to GO-CAM annotation guidelines, ontology lookup, and tools for reading models directly.

Running a literature-backed annotation project

To annotate a pathway from primary literature (verified claims doc, comments, pathway diagram), use the /gocam-claims-pipeline skill.

1. Create a process folder under gocam_models/, lowercase and hyphenated:

   PROCESS=ampar-endocytosis   # your pathway name
   mkdir -p gocam_models/$PROCESS/{literature,noctua,datamine,agent-output}

2. Add PDFs to gocam_models/$PROCESS/literature/, named <PMID>.pdf (e.g. 15664178.pdf). For Stage 2 (extending an existing GO-CAM), also drop the OWL/TTL into noctua/.

3. Run the pipeline: ask Claude "Run the GO-CAM claims pipeline for ampar-endocytosis". The pipeline reads instructions from workflow/ and writes outputs to gocam_models/$PROCESS/agent-output/.

4. Review outputs. The primary deliverable is agent-output/expert_validation_claims.md. Optional follow-ups (separate sessions): /validate-claims, /gemini-dual-review, /blind-review-claude, /blind-review-gemini.

See CLAUDE.md for the full skill list and project layout. Pipeline phases live in workflow/.