Build apps with the Claude API or Anthropic SDK. TRIGGER when: code imports `anthropic`/`@anthropic-ai/sdk`/`claude_agent_sdk`, or user asks to use Claude API, Anthropic SDKs, or Agent SDK. DO NOT TRIGGER when: code imports `openai`/other AI SDK, general programming questions.
Run automated experiment loops to optimize algorithms, parameters, or prompts. Two modes: 'grid' for systematic parameter sweeps with Optuna TPE + WilcoxonPruner, 'creative' for Karpathy-style LLM-proposed changes. Use when: optimizing algorithm parameters, tuning thresholds, prompt optimization, ablation studies, or any 'modify → measure → keep/discard' workflow. Triggers on: /autoresearch, 'optimize these parameters', 'run experiments', 'karpathy loop', 'ablation study', 'parameter sweep', 'tune this algorithm'.
Generate interactive HTML evaluation pages for collecting human ground-truth judgments on structured data from any project. Use when a project needs calibration data, quality audits, inter-rater evaluation, algorithm tuning, or ground truth labeling. Triggers on: /calibration-eval, 'evaluate these', 'need ground truth', 'calibration data', 'quality audit', 'rate these', 'human evaluation', 'label these items'.
Adaptively map what a user knows about a topic via an interactive HTML assessment page, then generate a personalized HTML explainer. Uses limbic.amygdala's knowledge_map module for entropy-maximizing probes and Bayesian belief propagation. Can load existing knowledge graphs or generate new ones from topic descriptions. Triggers on: /knowledge-probe, 'probe my knowledge', 'what do I know about', 'map my knowledge', 'knowledge probe'.
Generate a detailed, pedagogical HTML walkthrough of a PR or feature branch. Use when the user wants to understand, document, or share how a PR works — covering architecture, data flow, design decisions, tradeoffs, and extension patterns. Triggers on requests like "walk me through this PR", "explain this branch", "write up how this feature works", "create a walkthrough".
Write Playwright e2e tests with semantic selectors and accessibility-first assertions. Use when creating or modifying e2e tests (*.spec.ts files using Playwright).
Write README documentation focused on deep understanding and mental models. Use when asked to document a package, write a README, or create documentation for a library.
Reviews React code for correct useEffect usage. Use when reviewing React components, checking for useEffect anti-patterns, or deciding whether useEffect is appropriate for a given use case.