autoalphafold3-trial-submit

name	autoalphafold3-trial-submit
description	Use when validating and submitting exactly one auto-AlphaFold3 AutoFoldTrial JSON through the approved local orchestrator command, rejecting direct Modal calls, locked benchmark paths, hidden-validation search, missing hypotheses, and multi-trial submissions.

auto-AlphaFold3 Trial Submit

Use this skill after a research hypothesis exists and before any experiment is run. This skill is a narrow gate: validate one AutoFoldTrial JSON and submit only through the local orchestrator.

Workflow

1. Read references/trial-schema.md.
2. Confirm there is exactly one trial.
3. Confirm the trial has one hypothesis, one diagnostic target, one move family, allowed paths, a valid budget, and safe sampler semantics.
4. If valid, return the command: python -m autoalphafold3.agent submit trials/<id>.json
5. If invalid, refuse with reasons and do not provide any execution command.

Hard Rules

• Do not call modal run, modal.Function.from_name, .spawn, .remote, or modal.Sandbox.create.
• Do not use hidden validation during broad search.
• Do not accept edits to scorer, result parser, locked data, baseline ledger, validation splits, preprocessing outputs, Modal app, orchestrator, Volumes, GPU type, timeout, max_containers, or cost caps.
• Sampler-only trials must use a frozen checkpoint and label-free selection.
• The orchestrator is the only component allowed to spawn Modal work or append the canonical ledger.

Output Contract

For valid input:

VALID: true
TRIAL_ID: <id>
COMMAND: python -m autoalphafold3.agent submit trials/<id>.json

For invalid input:

VALID: false
REASONS:
- <reason>
COMMAND: <none>

References

• Read references/trial-schema.md for required fields, allowed values, and rejection checks.