Skip to content

LOOP.md

Latest commit

 

History

History
87 lines (66 loc) · 4.04 KB

LOOP.md

File metadata and controls

87 lines (66 loc) · 4.04 KB
Raw

You are inside the OpenAMP Foundry repo.

First pull. Read AGENTS.md, CLAUDE.md, MISSION.md, README.md, and the canonical docs below before choosing work:

  • docs/README.md — documentation routes and source-of-truth rules
  • docs/engineering/ARCHITECTURE.md — system architecture and extension points
  • docs/research/PLAN.md — current execution plan
  • docs/research/ROADMAP.md — current project state and milestones
  • docs/evidence/METRICS_CURRENT.md — measured evidence and known weaknesses
  • docs/trust/SAFE_SCOPE.md — safe-scope boundary

These files replace the retired flat paths docs/ARCHITECTURE.md, docs/PLAN.md, docs/ROADMAP.md, docs/METRICS_CURRENT.md, and docs/SAFE_SCOPE.md. Run one focused loop. When done: create PR, review, merge.

Mission: Move OpenAMP one real step closer to: “Open AI pipeline discovers a new antimicrobial peptide family, independently validated in lab tests, with full reproducible evidence trail released for scientific review.”

Goal: Build an open, safety-first wet-lab compression engine for AMP discovery: fewer, smarter experiments, honest learning from every batch. Make the repo more agent-friendly: clearer context, guardrails, architecture, tests, docs, workflows.

Rules:

  • Preserve safety, dry-lab, evidence-first constraints.
  • Never overclaim: computational outputs are not biological proof.
  • Prefer highest-leverage unfinished bottleneck.
  • Improve benchmark honesty, reproducibility, evidence certificates, reviewability, calibration readiness, or selection defensibility.
  • Reduce drift, duplication, stale docs, hidden assumptions, brittle workflows, architecture confusion.
  • Do not present future capability as current.
  • For virtual assays/simulation/active learning: include uncertainty, limits, baselines.
  • Preserve failures, caveats, rejected ideas, negative evidence.
  • Do not weaken novelty, toxicity, hemolysis, or dual-use safeguards.
  • Docs operational; code smallest useful slice.
  • Self-improvement useful, not ceremony.

Lenses: structure; architecture; agent usability; roadmap freshness; evidence plumbing; technical debt; verification surface.

Procedure: A. Inspect repo state, docs, tests, roadmap, drift. B. Pick highest-leverage bottleneck. C. Classify it: scientific credibility; benchmark honesty; candidate selection quality; wet-lab readiness; virtual-assay scaffolding; repo structure/architecture; agent workflow/self-improvement; documentation/handoff clarity. D. Explain why it matters now and outranks nearby work. E. Implement next meaningful slice end-to-end. F. Run verification if possible. G. Update docs/prompts/architecture/roadmap if state or execution quality changed. H. Leave repo easier for next loop. I. Report changes, evidence, uncertainty, and likely next loop.

Priority:

  1. Benchmark honesty / preventing self-deception.
  2. Evidence certificates, reproducibility, selection defensibility.
  3. Assay-ready batch or wet-lab feedback ingestion.
  4. Virtual-assay/calibration/active-learning scaffolding without theater.
  5. Repo structure, architecture, automation, handoff.
  6. Clarity for reviewers, collaborators, lab partners.

Prefer repeated leverage, truth-finding, lower friction, and work matching repo stage.

Allowed self-improvement: tighten structure/naming; improve runbooks/checklists/templates/logs; clarify interfaces/schemas/contracts; add tests/checks/scripts/CLIs; improve status tracking; consolidate docs.

Avoid: vague process; aspirational docs; extra ceremony; unjustified complex infra; PM theater.

Output:

  1. Chosen bottleneck + class.
  2. Why it matters.
  3. Changes made.
  4. Verification run.
  5. Risks / uncertainties.
  6. Recommended next loop. If self-improvement changed:
  7. Future loop benefit — how repo is easier, safer, faster, or clearer.

Before finishing, answer:

  • Is repo more truthful?
  • Easier to continue?
  • Architecture clearer?
  • Next bottleneck more visible?

If blocked: resolve, narrow, or produce best safe artifact to unblock next loop. Each step end we create a gh pr, then review, and merge. Then repeat this loop again - NEVER STOP IMPROVING