Claude skills should split prose, scripts, and tools instead of keeping everything as text
socialwithaayan · x · 2026-07-24
The author argues that in Claude skills, the important distinction is not whether a capability is written as prose or code, but whether it should be expressed as an instruction, a script, or a tool call.
Their rule of thumb is:
- Keep prose for judgment, tone, and context-sensitive adaptation.
- Move deterministic work to scripts: calculations, format conversions, file operations, validation checks, and any task that should always produce the same output.
- Use real tools/connectors when the skill needs to do actual work rather than merely describe how work should happen.
The broader lesson is that skills rot when they stay as instructions for things that should have been automated.
More from coding & agent
- How one developer uses ChatGPT Voice as a hands-free control panel for Codex — Dimillian · 2026-07-24
- A practical breakdown of 10 AI agent types, from reactive to multi-agent systems — goyalshaliniuk · 2026-07-24
- TRL and OpenEnv now let you train agents inside your own coding harness — ben_burtenshaw · 2026-07-24
- TRL plus OpenEnv can now train agents inside custom harnesses like OpenCode and Claude Code — ben_burtenshaw · 2026-07-24
- Favro-MCP ships a 32-tool Go server with one-line install for 13 AI clients — --lael-- · 2026-07-24
- Do automated evals work? A comparison of LLM tools on app traces — HamelHusain · 2026-07-24