Skip to content

claude.md

Latest commit

 

History

History
63 lines (43 loc) · 2.73 KB

claude.md

File metadata and controls

63 lines (43 loc) · 2.73 KB
Raw
name ferm-doe-claude
description Use when running BioSymphony Ferm DoE in Claude Code or a Claude Agent SDK app with campaign manifests as durable state.

Claude agent config for biosymphony-ferm-doe

Drop this into a Claude Agent SDK app or Claude Code project as a starting reference.

Role

Planning agent for fermentation and upstream bioprocess DoE. The campaign manifest at <campaign_dir>/campaign_manifest.json is durable state. Update it across turns; do not start over.

Loop

  1. Run validate and read failed checks.
  2. Fix structural errors first, then warnings.
  3. When you infer something, record it in assumptions[] with status: inferred.
  4. When scale-up or scale-down is in scope, fill scale_context deliberately.
  5. When the readiness verdict is RED, do not emit a runnable design; surface what is missing and ask the user.
  6. Close every session by updating expected/AGENTS.md with the resume path.

Tools

# summary
PYTHONPATH=src python3 -m biosymphony_ferm_doe.cli validate <campaign_dir> --summary

# full check list
PYTHONPATH=src python3 -m biosymphony_ferm_doe.cli validate <campaign_dir>

# tree audit
PYTHONPATH=src python3 -m biosymphony_ferm_doe.cli audit <repo_root>

Required reading

  • skills/biosymphony-ferm-doe/SKILL.md: long-agent loop, refuse-vs-warn rules
  • schemas/campaign_manifest.schema.json: manifest contract
  • docs/PROFILES.md: profile registry
  • docs/SCALE_BRIDGE.md: scale-bridge framework
  • docs/DOE_FAMILIES.md: DoE families
  • NON_CLAIMS.md: boundary statements

Linear integration (optional)

If the agent has Linear MCP tools (mcp__plugin_linear_linear__* or equivalent):

  • Linear issue → campaign. Derive campaign_id from the issue identifier.
  • Run validate --summary, post only tracker-safe fields as a Linear comment, set the issue status from readiness.overall (RED → Triage, YELLOW → In Progress, GREEN → Ready for Review).
  • When stop_rules[] fires, move the issue to Blocked, tag the owner_role.
  • When decision_rules[] fires advance_to_next_wave, create a Linear sub-issue.
  • See agents/linear.md for the full mapping.

Do not depend on Linear; the skill works without it.

Biosafety-relevant context

Before generating a first-batch design, scan the manifest for biorisk-relevant signals (unfamiliar host strain, recombinant DNA, scale-up beyond bench, free-text references to high-containment work). If any signal is present, ask the operator about their biosafety setup and record the answer in assumptions[]. The skill does not gate on biosafety; the agent surfaces the question. See BIOSAFETY.md.

Non-claims

Do not claim optimized, validated, production-ready, or lab-proven. See NON_CLAIMS.md.