6 key insights from Anthropic's skills guide: progressive disclosure is king
eherrerosj · x · 2026-10-14
eherrerosj distilled six key insights from Anthropic's skills guide:
- Name skills with gerunds (e.g. processing-pdfs) for clarity and consistency
- Keep references one level deep — chained links may only get previewed (first 100 chars)
- Add a table of contents since agents read first 100 lines to judge relevance
- Evaluation-driven dev: run Claude without the skill first, note failures, build test scenarios and a baseline, then write only what fixes the gaps
- Treat the description as the product: it's always loaded and decides if Claude picks your skill out of 100+; write it in third person with usage triggers
- Progressive disclosure is king: only name+description loads at startup; keep SKILL.md under 500 lines and push the rest to external files
More from coding & agent
- Self-built Agent Atlas dashboard shows 64 agents running in parallel — airesearch12 · 2026-10-14
- Swapping to Antigravity CLI Lifted Gemini 4 Argon to Near Opus 5.5 in Real-Work Bug Benchmark — PawelHuryn · 2026-10-14
- Schema retrieval judged by recall alone misleads agent builders: author proposes per-token coverage and cost metrics — burkov · 2026-10-14
- 654 Real-World Failures: First Large-Scale Study Maps How RAG Configuration Issues Break Systems — _reachsumit · 2026-10-14
- Impatient for JetKVM Mini, dev buys LuckFox KVM to let agents control his Mac Mini — TheZachMueller · 2026-10-14
- Airbnb case study: portable agent skill hits 99.2% procedural accuracy in 1-2 engineer days — _reachsumit · 2026-10-14