# Navigating Claude Code: Skills That Actually Fire > [!question] Lingering Questions > - Skill precedence, description budgets, and version thresholds are time-sensitive — recheck against current docs. > - "Auto-invocation remains a semantic judgment" per the deep-read — the article doesn't benchmark how often good-vs-bad descriptions actually change firing rate; this is presented as heuristic, not measured. > [!tip] Article Ideas > - "The description needs work — not the instructions" reframes a common skill-authoring mistake (over-investing in the body, under-investing in the trigger). Worth a short practical checklist for anyone maintaining a large personal skill library (directly relevant to Courtney's own ~/.claude/skills setup). > [!link] Cross Reference > - Part 10 of Oleg Efimov's 10-article HackerNoon series; see [[3. Resources/Literature Notes/Navigating Claude Code - The Full Workflow]] for series context — this is the final installment. > - [[3. Resources/Permanent Notes/A Declared Capability Costs You Every Turn Not Every Use]] — same context-cost logic applied to skill descriptions rather than MCP tool schemas ("keep the skill inventory focused because descriptions... consume context") > [!quote] Vocabulary > - **Explicit invocation** — calling a skill by name/command rather than relying on the model to infer it should fire from the task description; recommended for destructive or high-consequence workflows ## Bibliographic capsule Article, HackerNoon, 2026-08-04. Tenth and final installment of Oleg Efimov's Claude Code series. ## Author's thesis (Adler Q1+Q2) **What is the article about?** What makes a Claude Code skill reliably auto-invoke at the right moment, versus sitting unused or firing incorrectly. **What is being argued?** Skill reliability depends more on concise, task-triggering descriptions and distinct names than on how thorough the skill's internal instructions are — write descriptions as task triggers, front-load the use case, reserve explicit invocation for destructive/high-consequence workflows where silent auto-invocation is too risky, and keep the total skill inventory focused since every loaded description costs context. ## Key arguments (≤3) 1. **Write descriptions as task triggers, front-load the use case** — the description is what Claude reads to decide whether to load the skill; it needs to match how a task is actually phrased, not summarize the skill's contents. 2. **Use explicit invocation for destructive or high-consequence workflows** — auto-invocation is a probabilistic judgment; for anything with real consequences if it fires wrongly (or doesn't fire when needed), require an explicit call instead. 3. **Keep the skill inventory focused** — every skill's description (and, once loaded, its instructions) consumes context whether or not it ends up being the one used. ## Claim atoms ### The description is the trigger, not a summary > [!quote] ID:rw_01kz7sts1k29515re1z4gw6232 > The description is what Claude reads to decide whether to load this skill automatically Where/why: the article's central reframe of what a skill description is for. ### High-consequence workflows need explicit invocation, not auto-fire > [!quote] ID:rw_01kz7stvz7cv0cv7jqq1xw83wk > There is no warning. Where/why: names the risk of relying on probabilistic auto-invocation for destructive actions. ### A weak description is the usual failure, not weak instructions > [!quote] ID:rw_01kz7stwn70mhp5a5vystjxjck > description needs work — not the instructions Where/why: the article's diagnostic heuristic for a skill that isn't firing when expected. ## Permanent candidates - **A skill's description is the interface that decides whether it's used at all, independent of instruction quality** — genuine, distinct claim worth its own note; directly actionable for maintaining a large personal skill library. ## Processing checklist - [x] Checked existing Permanents — no close match for the description-as-trigger claim specifically; created new - [x] Connect into the relevant MOC — [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]], [[3. Resources/MOCs/Personal Knowledge Management MOC]] ## Links - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — agent-tooling design practice - [[3. Resources/MOCs/Personal Knowledge Management MOC]] — directly relevant to maintaining a large personal skill library as a knowledge-retrieval system - Metadata: [[3. Resources/Readwise/Articles/Navigating Claude Code Skills That Actually Fire|Navigating Claude Code - Skills That Actually Fire]]