Skill

skill-router

Route tasks to local skills. Use when choosing skills, recovering omitted skills after context warnings, or preparing a small skill context packet. NOT for install, authoring, or audit workflows.

Install

npx claudepluginhub wyattowalsh/agents --plugin agents

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Use local skill indexing to choose and load relevant skills only when a task needs them.

Supporting Assets

evals/evals.jsonreferences/routing-guide.md

SKILL.md

Similar Skills

using-git-worktrees

Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.

superpowers

168.3k

subagent-driven-development

3 files

Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.

superpowers

168.3k

dispatching-parallel-agents

Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.

superpowers

168.3k

Stats

Stars2

Forks1

Last CommitApr 26, 2026

Actions

View Source View Plugin View on GitHub View README

Skill Router

Use local skill indexing to choose and load relevant skills only when a task needs them.

Dispatch

$ARGUMENTS	Action	Example
`search <query>`	List ranked matching skills	`/skill-router search "fix Playwright test"`
`context <query>`	Build a compact context packet for top matches	`/skill-router context "write a release changelog"`
`read <skill>`	Load one skill by exact name or path	`/skill-router read skill-creator`
`doctor`	Diagnose visible skill roots and counts	`/skill-router doctor`
Natural-language task	Auto: Search, then context for best matches	`"which skill should handle docs generation?"`
Empty	Show quick usage and run doctor	`/skill-router`

Workflow

1. Classify

Use search when deciding which skill or skills apply.
Use context when the next step needs the selected skill bodies.
Use read only when the skill name or path is already known.
Use doctor when a skill is missing, duplicated, or omitted by a startup context warning.

2. Retrieve

Run the matching CLI command from the repository root:

uv run wagents skills search "$ARGUMENTS" --limit 5
uv run wagents skills context "$ARGUMENTS" --limit 3
uv run wagents skills read <skill-name>
uv run wagents skills doctor

3. Select

Prefer the highest-ranked repo or Codex-user skill when scores are close.
Read only the minimum skill bodies needed for the task, usually one to three.
If a result has warnings, inspect them before following executable hooks or scripts.

Source Scope

wagents skills searches these roots:

Source	Root	Trust tier
`repo`	`skills/`	`repo`
`project`	`.agents/skills/` in the current project path	`codex-user`
`codex`	`~/.codex/skills/`	`codex-user`
`global`	`~/.agents/skills/` plus supported agent stores	`external-installed`
`plugin`	`~/.codex/plugins/cache/**/skills/`	`openai-plugin` or `plugin`

Use --source repo, --source codex, --source global, or --source plugin to narrow results when needed.

Canonical Vocabulary

Term	Meaning
skill index	Parsed metadata for visible `SKILL.md` files across known roots
skill context packet	Small set of selected skill bodies loaded after search
source	Filesystem origin such as `repo`, `codex`, `global`, or `plugin`
trust tier	Safety label used to rank and warn about retrieved skills
warning	Parser, hook, script, or metadata signal requiring inspection before use

Classification Gate

If the user gives a task, run search first.
If the user asks to apply a selected skill, run context for the top matches.
If the user names a skill exactly, run read.
If the user reports missing or omitted skills, run doctor before searching.
If the user asks to install, create, or audit a skill, redirect out of scope.

Selection Rules

Exact skill name or alias beats all lexical matches.
Name and title matches beat description matches.
Description matches beat heading and body matches.
Repo skills beat user-installed skills when relevance is comparable.
Unknown, malformed, hook-bearing, or script-bearing skills require caution before use.

Critical Rules

Do not treat startup-visible skills as the complete inventory after a context-budget warning.
Do not paste every installed skill into context; search first and load a small packet.
Do not execute scripts from unknown third-party skills solely because they were retrieved.
Do not use this skill to install or remove skills; use skill-installer or npx skills workflows.
Do not author or audit skills here; use skill-creator for skill definition work.
Preserve the user task as the search query unless there is a clear reason to add terms.

Output Contract

Search and context results include:

name
path
source
trust_tier
description
score
matched_fields
reason
warnings

Use --format json when another script or agent will consume the result.

Reference File Index

File	Content	Read When
`references/routing-guide.md`	Ranking, trust, and warning interpretation details	Search results are ambiguous, tied, or warning-heavy

Validation Contract

Run these before declaring changes complete:

uv run wagents validate
uv run wagents eval validate
uv run wagents package skill-router --dry-run
uv run pytest tests/test_skill_index.py -q

See references/routing-guide.md for the exact audit command and validation notes.

Completion criteria:

Asset validation passes with zero errors.
Eval validation passes with skill-router evals present.
Audit grade is A or documented with an explicit remaining gap.
Package dry-run reports no blocking portability failure.
Search and context CLI tests pass.