From latestaiagents
Design skills that fire at the right moment — neither over-eager (noise) nor under-eager (silent). Covers activation specificity, trigger phrases, disambiguation between overlapping skills, and debugging activation. Use this skill when multiple skills could fire on the same query, a skill never fires, or a skill fires too often. Activate when: skill won't activate, skill over-activates, overlapping skills, skill triggers, skill selection, skill disambiguation.
npx claudepluginhub latestaiagents/agent-skills --plugin skills-authoringThis skill uses the workspace's default tool permissions.
**A skill that fires at the wrong time is as useless as one that never fires. Design activation like you'd design a cron expression: specific, testable, boundaried.**
Guides creating skills from scratch, modifying existing ones, or optimizing trigger descriptions via intent capture, SKILL.md structure, testing, and progressive disclosure.
Create, improve, and test SKILL.md files to extend Claude Code with project-specific knowledge and reusable workflows.
Creates new Claude Code skills from scratch following best practices for structure, naming, frontmatter, progressive disclosure, reference organization, and tool scoping. Use for building skills or converting slash commands.
Share bugs, ideas, or general feedback.
A skill that fires at the wrong time is as useless as one that never fires. Design activation like you'd design a cron expression: specific, testable, boundaried.
The model matches your skill against user queries using:
Invest most in (3) and (4).
Skill fires when user pastes or describes an error:
description: |
Decode Java stack traces and suggest root causes.
Use this skill when user shares a stack trace, exception, or runtime error in Java.
Activate when: Exception, NullPointerException, stack trace, Caused by, Java error, runtime crash.
Include the literal tokens that appear in errors. Users paste them verbatim.
Skill fires when user expresses a goal:
description: |
Refactor code to extract methods, rename, and improve structure without changing behavior.
Use this skill when user asks to refactor, clean up, or restructure existing code.
Activate when: refactor, clean up, extract method, rename, restructure, DRY up, improve this code.
Match imperative verbs and goal phrases.
Skill fires when the user mentions a specific tool:
description: |
Configure Next.js App Router — server components, server actions, middleware, caching.
Use this skill when the user is working with Next.js App Router, specifically app/ directory routes.
Activate when: Next.js, App Router, server component, server action, app/ directory, use server.
Name the technology and its key concepts; matches work well.
Skill fires during a known phase of work:
description: |
Pre-release checklist for a production deploy — changelog, version bump, migration safety, smoke tests.
Use this skill when user is about to deploy to production, cut a release, or tag a version.
Activate when: deploy to prod, release, version bump, production release, pre-release check.
Workflow skills need to be clearly scoped to that phase or they fire constantly.
If two skills could both fire on "help me write a test", they compete. Differentiate in the description itself.
On "write a test for MyComponent", both fire. Model picks randomly.
The "NOT for..." line is a powerful routing hint.
Symptom: skill fires on nearly every query. Cause: description too broad.
description: |
Help with coding tasks.
Activate when: code, programming, development.
description: |
Help refactoring existing code for readability — NOT for new code generation, bug fixing, or debugging.
Use this skill when user specifically asks to refactor, clean up, or restructure.
Activate when: refactor, clean up, extract method, DRY up, improve readability.
"NOT for X" sentences are explicit guards. Use them.
Symptom: skill has clear use case but never fires. Cause: description misses the user's vocabulary.
Audit real queries users send. Add their exact phrasings:
Activate when:
- merge conflict, resolve conflicts # jargon
- CONFLICT markers, <<<<<<< HEAD # literal tokens
- git says there's a conflict # natural phrasing
- merge failed, can't merge # error paraphrases
- rebase stopped, rebase conflict # related workflows
Mix jargon, error strings, and natural phrasings.
Some patterns explicitly should NOT activate the skill. Call them out:
description: |
Write SQL migrations safely.
Use for: ALTER TABLE, new columns with defaults, adding indexes on prod tables.
DO NOT use for: read-only queries (use sql-query-helper), ORM model changes, new tables from scratch.
Activate when: ALTER TABLE, add column, migration, safe migration, Postgres lock, online DDL.
Explicit negatives reduce competition with nearby skills.
See the skill-testing companion skill. Briefly:
help, handle, manage. Match too broadlytests, bugs, config