Install
$ agentstack add skill-mahdisanagostar-skill-finder-skill-finder ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
Skill Finder
Mission
Use this as Chef's automatic routing layer.
Run it:
- right after
$chef-indexfor Codex work - before choosing specialist skills in Claude work
- again whenever the task changes shape, artifact, or risk
The goal is not to list every possible skill. The goal is to engage the fewest skills that materially improve the work.
Fast Path
Do not force routing when the task is already small and obvious.
Skip specialist selection when:
- the user asks for a trivial shell action or one-command lookup
- the user names one exact skill or plugin route and no support changes execution
- the current step is a tiny follow-up inside an already-correct skill lane
Default Behavior
For every non-trivial task:
- honor explicit skill, plugin, or app requests first
- classify the task
- pick one primary skill
- add at most one execution support and one verification support
- explain the choice in one short line
- re-run the check if scope shifts
If no specialist skill clearly helps, say so and continue without forcing one.
Classification Pass
Look at seven things first:
intent
build, review, debug, research, document, design, present, automate
artifact
code, UI, deck, doc, spreadsheet, repo config, workflow, knowledge base
risk
security, architecture, multi-module change, user-facing polish, external publish
medium
terminal, browser, slides, images, docs, video
freshness
whether direct graph read, live browsing, or current repo state matters
host
Codex runtime, Codex plugin, shared Chef skill, Claude skill
explicit cues
user-named skills, plugins, tools, output formats, or platforms
Use the matrix in references/selection-matrix.md when the answer is not obvious.
Routing Rules
1. Prefer one strong specialist
Pick narrowest skill that squarely matches task.
Good:
- frontend request ->
frontend-design - presentation request ->
talkcraft - skill creation request ->
skill-creator
Bad:
- frontend request ->
frontend-design,design-sprint,ux-heuristics,refactoring-ui
2. Explicit request wins
If user names skill, plugin, or tool surface, start there unless it clearly conflicts with task.
Examples:
- asks for GitHub PR triage ->
github:github - asks for Notion capture ->
notion:notion-knowledge-capture - asks for PowerPoint deck ->
slides - asks for OpenAI API guidance ->
openai-docs
3. Prefer host-native specialists before generic overlap
When current host has a stronger native route, prefer it over a broader fallback.
Examples:
- editable
.pptxbuild ->slides, not onlytalkcraft - spreadsheet file work ->
spreadsheetorExcel, notexcel-mcp-server - GitHub Actions CI fix ->
github:gh-fix-ci, not generic review flow - Slack send or draft ->
slack:slack-outgoing-message, not plain messaging advice
4. Add support only when it changes execution
Supporting skills should change how the work happens, not just sound relevant.
Common support patterns:
- code change + isolation need ->
using-git-worktrees - code change + review risk ->
code-reviewer - code change + security concern ->
secure-code-guardian - codebase question + repo context ->
graph-first-retrieval - deck build + weak narrative ->
talkcraft - editable artifact + final verification need -> medium-specific verification skill
Treat support lanes as:
execution support
worktree, graph retrieval, implementation workflow
verification support
code review, security review, browser verification
Do not add two supports from same lane unless user explicitly asks.
5. Avoid overlap twins
Never choose two skills that solve same layer of problem unless user explicitly asks for both.
Common overlaps:
slidesandPowerPointspreadsheetandExcelplaywrightandplaywright-skillfrontend-designandbuild-web-apps:frontend-skillcode-reviewerandgithub:gh-address-comments
6. Respect stage changes
Re-route when the task moves from one stage to another.
Examples:
- idea -> implementation: add
feature-forgeorusing-git-worktrees - implementation -> review: add
code-reviewer - notes -> deck strategy: switch to
talkcraft - static design -> interactive artifact: switch to
web-artifacts-builder
7. Respect host and project reality
- prefer enabled Chef skills listed in
AGENTS.mdorCLAUDE.md - prefer Codex runtime or plugin skills when current host exposes them
- prefer bundled local skills over hypothetical ones
- do not route to unavailable skills
8. Keep the set small
Hard cap:
1primary skill0-2supporting skills
If you want more than three, the routing pass failed.
Output Format
When routing matters, announce the decision briefly:
Using $skill-finder: primary = $frontend-design; support = $using-git-worktrees because this is a UI build with implementation work.
Using $skill-finder: primary = $slides; support = $talkcraft because user needs an editable deck and stronger story structure.
Using $skill-finder: no specialist added because this is a trivial shell task.
For Claude-host work, use the same pattern without $ if needed.
Re-Route Triggers
Run the routing pass again when:
- the user changes the deliverable
- the task moves from planning to implementation
- the work becomes security-sensitive
- the work becomes presentation- or media-heavy
- the repo reveals a better specialist than the first guess
Always-On Chef Role
Inside Chef, this skill acts as the default skill broker.
chef-indexestablishes Chef contextskill-finderchooses specialist skills- specialist skills drive the actual work
Do not skip straight from chef-index to execution on substantial tasks.
Reference Files
references/selection-matrix.md
Primary routing table from task type to skill choice.
references/routing-rules.md
Guardrails, anti-patterns, and re-routing rules.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: mahdisanagostar
- Source: mahdisanagostar/skill-finder
- License: MIT
- Homepage: https://github.com/mahdisanagostar/chef
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.