Skills written entirely by the model are useless, and stacking skills taxes agent context
HamelHusain · x · 2026-10-26
Hamel Husain amplifies a discussion on agent skills: a skill written entirely by the model with no human input is one you don't need — value comes from injected domain knowledge. On whether consuming many skills in one session dirties context, dexhorthy's take: 5 steps in a single session is fine; you only need subagents/compaction when skills must stack (earlier instructions still honored later) or hundreds of skills must stay active, since each skill takes a share of attention.
More from coding & agent
- Traks: free MIT-licensed self-hosted analytics on Cloudflare with an MCP server for agents — moeinteractive · 2026-10-26
- Open-source AI design skills for Claude Code and Cursor earn 5.1k stars — moeinteractive · 2026-10-26
- RExBench: best coding agent solves only 33% of autonomous research extension tasks — najoungkim · 2026-10-26
- Remotion ships official Agent Skills for Claude Code, Codex and Cursor — lmoroney · 2026-10-26
- Making motion graphics with Claude and Remotion: video as editable React code — lmoroney · 2026-10-26
- Ruff creator on quality: 'Caring is the moat' — by the time data shows it, it's too late — charliermarsh · 2026-10-26