← ClaudeAtlas

competition-model-quality-freezelisted

Resume or freeze a contest-modeling checkout when the user says “继续”, asks for quality verification, or needs paper/submission gates kept separate from local results.
ychenfen/claude-agent-skills-collection · ★ 0 · AI & Automation · score 78
Install: claude install-skill ychenfen/claude-agent-skills-collection
# Competition Model Quality Freeze ## When to use Use for a competition-modeling project with existing code, results, handoffs, or a proposed paper/submission. Use it to recover context, verify a checkpoint, or prepare a truthful freeze. Do not use it to claim submission, upload, or final scientific conclusions from an incomplete pilot. ## Inputs / context to gather 1. Confirm `$ARGUMENTS`, its actual working subdirectory, and the boundary between official prompt/attachments/templates and editable implementation. 2. Read the current handoff, modeling log, manifest, requirements coverage, quality/freeze checklist, and result directory before changing code. 3. Inventory the current paper/PDF, figures, generation scripts, test commands, result CSV/JSON, archive, and previous hashes. Classify each as current evidence, historical baseline, or missing. ## Procedure 1. For “继续”, recover the active objective and gates from local artifacts; begin with a read-only status table. State explicitly whether the status is exploratory, locally verified, frozen, packaged, uploaded, or submitted. 2. Preserve official originals and create/identify a recoverable rollback artifact before model changes. Keep exploratory candidates isolated from formal runs. 3. Run the project’s tests, compilation/static checks, and domain assertions. For experiment pipelines, retain inputs, evaluator/version fingerprints, source provenance, and hashes; do not put cache-hit counters into scientific result CSVs