steering

Three-letter skills for agent wrangling.

See the main installation guide for Claude Code, Codex, and standalone Agent Skills setup.

Skills

• ask — Pause implementation and ask only the questions needed to produce a high-quality solution.
• cut — Make the response terser while preserving important reasoning, decisions, and next steps.
• dig — Revisit superficial or under-verified agent work by investigating its claims, assumptions, evidence, and missing edge cases.
• dyj — Stop punting and complete the user's requested task. Use after the model evades, defers, or stops before doing work it can perform.
• lfg — Implement the requested direction immediately using reasonable defaults and no unnecessary follow-up questions.
• lmk — Spawn a subagent to perform the research task, then report back with the answer.
• map — Explain the shape of an implementation, its key architectural decisions, and how its main pieces fit together.
• nah — Abandon the current direction, propose three genuinely different approaches, and recommend the strongest one.
• qna — Answer genuine questions directly without treating them as implied instructions to take action.
• wow — Rework a frontend with award-winning visual hierarchy, interaction design, polish, originality, and Webby-worthy quality.
• wtf — Explain the reasoning, evidence, assumptions, and tradeoffs behind the immediately preceding action or result.

Inspired by bro from dmmulroy/skills.