Dev argues agent skills are mostly slop, prefers a single master agent instead
teodorio · x · 2026-11-07
Developer teodorio argues that agent "skills" are mostly slop and useless: different wordings steer the model down different vector paths, and you can get stuck in good or bad token wells. His workaround is to skip skills and usually run a single master agent. The quip responds to a call for "god-tier prompts" — the idea that one densely informative sentence, where every verb, noun and adjective carries flavor, beats a pile of skill files.
More from coding & agent
- Claude Code Projects collaborative view resembles ChatGPT Spaces; Perplexity next — testingcatalog · 2026-11-07
- Anthropic reportedly building collaborative mode for Claude Code Projects — testingcatalog · 2026-11-07
- Context management is still early: correctness, latency and cost trade off — blaizedsouza · 2026-11-07
- odai framework adds governance and evidence checks to stop agents faking "done" — blaizedsouza · 2026-11-07
- BUI decouples Blender's GPU UI kernel into a standalone runtime: 18MB memory, 12ms cold start — Obijuan_cube · 2026-11-07
- Parcha on Agent Sandboxes: Lessons from Building Agents Since 2023 — blaizedsouza · 2026-11-07