bnufw

Use when the user wants a Prisma-style dual-branch idea workflow that takes a required Markdown brief plus local paper notes, runs one Codex-native branch and one Gemini branch, then compares both outputs and selects a winner.

Use when the user wants to brainstorm research ideas, identify gaps, define a research question, or plan an early-stage project from a user-provided Markdown brief, local paper notes in `papers/**/*.md`, and optional online paper search when local evidence is insufficient.

Use when the user wants to generate or refine one research idea from a required Markdown brief and local Markdown paper notes stored in `papers/**/*.md`, following `prompts/generate_idea.md`.

This skill should be used when the user asks to "learn from Kaggle", "study Kaggle solutions", "analyze Kaggle competitions", "learn from code", "extract patterns", or mentions Kaggle competition URLs. Also covers extracting reusable patterns from coding sessions.

Generate a project-specific AGENTS.md from a user goal, then confirm before overwriting.

Build docs/mcp-tools.md from the active MCP servers and available tool functions.

Draft a repo-local plan using the plan skill template and optionally save it.

Write, configure, and run unit + integration tests using project-independent best practices; detect tooling; triage failures/flakes; keep a compact Testing State Capsule.

Generate one repo-grounded oral-grade deep-learning idea together with a paper-level method, a full experiment program, and the first formal batch.

Implement the active formal batch from .codex/active_idea.md, disclose the batch with request_user_input, and launch the formal run in screen.

Review one completed formal batch, then either improve the idea through multi-round clarification, abandon it, or finish it.

Research-only background queue for Codex CLI. Use when the user wants Codex to evaluate ideas from ideas/*.md, run long experiments under screen, verify results, refine attempts, and stop when the configured research target is met.