ashtonchew

Use when acting as the auto-AlphaFold3 autoresearch agent: read baseline metrics and Fold Cartographer diagnostics, choose one diagnostic target and one move family, propose or patch only the allowed AlphaFold3-lite model/training/sampler surface, and hand one trial to the orchestrator without touching Modal infrastructure or locked benchmark files.

Use when operating as a bounded auto-AlphaFold3 subagent worker that produces one proposal artifact for an assigned diagnostic target or move family, supports fast parallel hypothesis fanout, and never submits trials, calls Modal, edits locked files, appends ledgers, or integrates its own result.

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.

Use when interpreting auto-AlphaFold3 scorer diagnostics into exactly one Fold Cartographer failure target, preserving best_val_calpha_lddt as the primary objective and mapping local geometry, long-range topology, distogram-vs-3D, or stability/compute evidence to safe move-family suggestions.

Use when a task depends on current Modal Python APIs, migration guidance, deployed object lookup, or docs-backed runtime behavior.

This skill provides guidance for working with the Modal cloud platform. Use this skill whenever the user mentions Modal or has code that imports the `modal` SDK in Python, Go, or JavaScript. This skill should also trigger when the user needs to run Python code with vertical or horizontal scalability (e.g. batch jobs), needs access to GPUs (e.g. AI workloads including training and inference) or needs to run untrusted processes in a sandbox, since Modal serves these use cases well.

Build, evaluate, and train AI agents on RL environments with HUD. Use whenever someone wants to create an RL environment, benchmark, eval, or training task — for a coding, computer-use, browser, or robotics agent — or run and grade tasks across any model (Claude, OpenAI, Gemini, or open/self-hosted models). Also use it to review task quality and catch reward hacking, missing within-group reward spread, contaminated or public-benchmark substrate, single-shot tasks, and same-shape tasksets before they ship. Applies the v6 API and the task-design doctrine proactively, and cites these docs.