theam

Set up and launch an autonomous research project with Limina. Use when the user wants a persistent research agent workflow with hypotheses, experiments, findings, and review artifacts.

Build, review, refactor, or design software for readability, simplicity, modularity, testability, and safe changeability. Use for feature work, architecture decisions, API boundaries, naming, refactors, and code reviews across any language or stack.

Challenge the current Limina research direction and next-step plan. Use for step-back reviews, pre-commitment checkpoints, plateau reviews, contradictory findings, or pivot/continue/stop recommendations when evidence or framing may be weak.

Design, review, and conclude rigorous research experiments for Limina's H→E→F workflow. Use when creating or revising hypotheses, experiments, or findings; defining baselines, datasets, metrics, or stopping rules; comparing candidate methods; deciding whether a result is conclusive; or when a negative result may be caused by an invalid setup rather than a true method failure.

Map the AI/ML state of the art for a concrete technical problem. Use for exploratory research, mechanism-based literature search, benchmark mapping, and finding generalizable approaches. Not for simple fact lookup.

Generates data-driven reports about the project. Use for initial project reports or session summaries.

Build evaluation datasets for AIFindr agents by adding queries with expected responses extracted from a knowledge base. Use when asked to add queries to a W&B dataset, create evaluation test cases, or build ground truth datasets from documents. Requires W&B dataset name, project, query, and knowledge base path.

Comprehensive evaluation of AIFindr agents against their knowledge base. Use when asked to evaluate an AIFindr agent, test agent responses, run evaluations from a W&B dataset, or generate evaluation reports. Supports single queries, multiple queries, or full dataset evaluation. Generates markdown reports and XLSX files.

Execute comprehensive E2E tests for aifindr-knowledge-base application using agent-browser. Use this skill when asked to test the knowledge base app, verify UI functionality, test CRUD operations (tenants, promotions, documents, web URLs), test the publish workflow, or verify data ingestion in Weaviate. Creates a new project for test isolation, uses XLSX test data with images, verifies image uploads display correctly, and tests cascade delete of promotions when tenant is deleted.

Transcribe audio files with speaker diarization using local ML models. Works in both Claude Code and Cowork mode.