Browse Skills

Review a pull request for real bugs, regressions, and convention violations. The skill enumerates candidate issues across the diff, verifies each against the code, and returns structured line-anchored findings and a verdict.

Cross-references code, docs, and tests to decide if a diagnosed issue is an actual bug or deliberately intended behavior. Acts as a gating step to prevent unnecessary code changes.

Reproduce a bug in the public-facing part of a web app by booting a demo, driving public routes with a browser agent, and capturing screenshots plus a written transcript.

Reproduce EmDash bugs in REST handlers, CLI, MCP, migrations, or build tooling. Creates a failing vitest test or repro script to isolate backend issues without a browser.

Reproduce an EmDash admin UI bug by booting a demo, driving the admin with agent-browser, and capturing screenshots and a replayable transcript.

Implements a diagnosed bug fix in the EmDash CMS codebase, following project conventions, adding regression tests, running lint/typecheck, and staging the change for human review.

Traces reproduced software symptoms to their root cause in source code, identifying the specific file and line. Use after reproducing a bug to pinpoint the code responsible and propose a fix.

Build EmDash CMS plugins featuring hooks, storage, admin UI, API routes, and custom Portable Text block types. Use when scaffolding or implementing plugins or adding custom block types, admin pages, or content hooks.

Fine-tune language models using reinforcement learning with the TRL library. Supports SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training.

Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.

Guidance for LLM post-training with RL using Slime, a framework combining Megatron-LM for training and SGLang for rollout. Useful for training GLM models, custom data generation, and tight Megatron-LM integration for RL scaling.

SimPO is a reference-free preference optimization method for LLM alignment, outperforming DPO without a reference model. Use for simpler, faster training on preference data compared to DPO or PPO.