# Skill Descriptions Are Trigger Keywords Not Documentation ## Claim A skill's description field functions as a search trigger for the model, not a manual, because it loads into context on every turn whether or not the skill fires — so I write it as short, searchable trigger phrases and put explanatory content in the skill body instead. ## Evidence **Source**: [[3. Resources/Literature Notes/My AGENTS.md & SKILLS.md Breakdown]] ## Scope & conditions - Holds when: the agent harness always-injects skill/tool descriptions into context regardless of whether the skill is used — true of Claude Code and similar MCP/skill systems. - Breaks when: the harness only surfaces a description after some other selection step (so it isn't a standing tax), or the library is small enough that description length isn't a measurable context cost. ## Links - [[AI & the Future of Knowledge Work MOC]] — because applies-in: agent-harness context-cost design is part of this MOC's amplifier-thesis-and-harness thread — the description field is one more always-loaded cost to budget deliberately. - [[3. Resources/Permanent Notes/Context Budget Is Spent By Default Not By Exception]] — because extends: that note argues for minimizing what's always-loaded at all (lazy-load by default); this note is the design rule for the one field that stays always-loaded regardless — write it as keywords, not prose. <!-- atomicity: 1 idea ✓ --> - [[3. Resources/MOCs/Documentation & Technical Writing MOC]] — applies-in: the docs discipline map, "docs that agents read" cluster