genomeft-intake

name	genomeft-intake
description	Use when the user naturally asks to start, continue, monitor, or summarize a GenomeHarness campaign or sequential benchmark suite.

GenomeHarness Intake

Use plans/genomeharness_v15_plan.md as the active authority.

New Work

For a new campaign or suite:

1. Translate the user's natural language into model, benchmark/task, scope, budget, and resource policy.
2. Never pass bare nt as a benchmark. Use nt-suite for the NT benchmark family or a full task id such as nt/H2AFZ.
3. Run read-only:

genharness plan --dry-run --project <repo> --model <model> --benchmark <nt-suite|gb-suite|task_id> --scope <suite|campaign> --json

4. Show the preview: search mode full_l3_async, full-data search, 4 slots, UCT settings, deadline+drain, GPU policy, confirmation/final policy, and test isolation.
5. Wait for user confirmation before suite-init, preflight, or start.

Continue Or Status

For explicit continue/status:

genharness status --campaign campaigns/X --brief
genharness advance --campaign campaigns/X --wait

If the runner is inconsistent:

genharness recover --campaign campaigns/X

User Interface

The user should not need to copy shell commands. Execute the genharness tool calls yourself and report concise progress.

Token Budget

• Keep previews and status updates compact.
• Prefer genharness status --brief for progress checks.
• Do not paste full JSON reports, logs, or trial tables unless the user asks.