ClawBio

ClawBio is a Python package and skill library for reproducible, local-first bioinformatics workflows. Use this repo skill to operate the platform and choose its specialist skills; do not improvise scientific thresholds or treat an output as a clinical diagnosis.

Start here

• For install, clawbio list/run/upload, Python APIs, profiles, outputs, or replay metadata, read core-runner.
• For a natural-language biological request, an input file, or a multi-step chain, read domain-routing before choosing an alias.
• For MCP, messaging/web adapters, provider credentials, or nf-core/Nextflow, read pipelines-integrations.
• For creating or changing a skill, catalog registration, tests, demos, or INTENTS.json, read skill-authoring.
• For replay/checksum review, benchmark status, action contracts, privacy, or security filtering, read validation-safety.
• Read troubleshooting when the failure spans more than one route. Use provenance before deciding whether this graph is stale for a repository checkout. For a read-only package/API/CLI probe, run check_environment.py after reading its help.

Public setup

For ordinary users, install the public distribution with Python 3.11 or newer:

python -m pip install clawbio
python -c "import clawbio; print(clawbio.__version__)"
clawbio list

The optional MCP extra is clawbio[mcp]. It is deliberately separate from the base install. A source checkout may use the repository's documented uv sync workflow, but a published skill must not assume that checkout exists.

Platform operating rules

1. Identify the goal, input format/build, desired artifact, and whether the request is demo, local-file, or profile based. Inspect small headers locally; do not upload raw genomic data just to determine its type.
2. Route to the narrowest catalog entry. Catalog metadata distinguishes the registered CLI aliases from agent-readable-only SKILL.md contracts; do not invent a runnable alias for a spec-only skill.
3. Execute registered skills through clawbio run or the stable Python runner, not by bypassing the runner's per-skill flag allowlist. Use a fresh writable output directory and inspect structured results as well as exit codes.
4. Preserve the local-first boundary. External APIs, cloud datasets, MCP local file access, bots, credentials, containers, Nextflow, and large references are explicit opt-ins with separate failure and provenance handling.
5. Keep PASS, SKIP, FAIL, stale/expired action, and unavailable-backend states distinct. A successful import or demo does not prove a scientific method, external service, GPU, or clinical interpretation.
6. Reports must retain the package's research/education disclaimer. Do not give medical diagnoses or treatment advice from a ClawBio result.

Common entry points

clawbio list
clawbio run <registered-alias> --demo
clawbio run <registered-alias> --input sample.ext --output results
clawbio upload --input genotype.txt --patient-id PT001
clawbio run profile --profile profiles/PT001.json --output profile-results
clawbio mcp

The exact flags and virtual full-profile behavior belong to core-runner. Use domain-routing for aliases and chains, not this short list as a substitute for input inspection.

Freshness and scope

This graph is versioned against the source snapshot in repo-provenance.md. If the package version, public entry points, catalog, or major evidence paths changed, refresh the repo skill before relying on detailed claims. Review-only test cases and reports are kept outside this runtime directory; they are not inputs to ordinary ClawBio runs.