Task match skills

Task Match Skills — Discover Skills While You Work

Stop guessing which skills exist. Describe your task in natural language and Task Match Skills finds the right tools for you.

What It Does

1. Understands your needs — Extracts domain, tools, actions, and related concepts from your request.
2. Checks local skills — Finds already-installed skills that might help.
3. Searches the ecosystem — Queries the open agent skills registry with multiple keyword combinations.
4. Ranks & recommends — Presents findings by relevance tier with install commands.

---

⚠️ Mandatory Output Requirements

Every recommendation table MUST include these 5 columns — no exceptions:

Column	Content	Example
Skill	Skill name + install path	owner/repo@skill-name
Description	One-line summary of what it does	"GitHub PR and issue automation"
Why this helps	How it relates to the user's request	"Automates the deploy workflow you described"
Source	Tool name + abbreviated keyword	find-skills → "ci cd"
Install	Exact install command	npx skills add owner/repo@skill-name

Source column format: <tool> → "<first keyword used>"

• If found by multiple tools, list the primary one (see priority below)
• Keep the keyword short — just the first search term, truncated to ~20 chars

Do NOT omit the Source or Install columns. Users need to know where each skill came from and how to install it. A recommendation without these is incomplete and unactionable.

---

🔍 Search Tool Priority

find-skills (the skill from vercel-labs/skills) is the most recognized and widely-used discovery tool in the Agent Skills ecosystem. Treat it as the primary search channel whenever available.

Priority	Tool	When to use
🥇 Primary	find-skills skill	Always, if installed locally
🥈 Secondary	npx skills find	Always available, use to supplement
🥉 Fallback	openskills, load-skill, etc.	If detected and primary/secondary miss coverage

Source attribution when multiple tools find the same skill: always credit find-skills first. Example: if both find-skills and npx skills find return the same result, the source column shows find-skills → "keyword".

---

Trigger Sensitivity Configuration

This skill ships with disable-model-invocation: false (auto-trigger on). If you find it triggers too often, adjust in your agent's settings.

Sensitivity levels (choose one)

Level	Behavior	How to set
🔋 Aggressive (default)	Triggers on most messages; skips only chitchat, simple Q&A, typo fixes	Current setting
⚖️ Balanced	Triggers when user clearly describes a task or asks for recommendations	Add paths: "!**/*.fix" to frontmatter
🎯 Conservative	Only triggers when user asks about skills/tools/automation explicitly	Set disable-model-invocation: true

---

Workflow

Step 1 — Triage

Does the user's message fall into the skip list?

• Skip: Pure chitchat, simple fact Q&A, narrow typo/bug fix
• Process: Everything else

Step 2 — Analyze Intent

Extract from the user's request:

Dimension	Ask yourself	Example: "deploy my React app to AWS"
Domain	What tech area?	Web deployment, cloud infra
Tools	What tools mentioned/implied?	React, AWS
Action	What does user want to do?	Deploy, host
Related	What else might they need?	CI/CD, Docker, monitoring, SSL, DNS

Don't just search the user's exact words — reason about what else the task might involve. Ask yourself:

1. Upstream: what needs to happen before this task?
2. Downstream: what happens after?
3. Toolchain: what tools typically work with the ones mentioned?
4. Adjacent domains: what related areas might have specialized skills?

Example — user says "deploy my React app to AWS":

• Upstream: CI/CD pipeline to build → search "ci cd github actions"
• Downstream: monitoring after deploy → search "monitoring logging alerting"
• Toolchain: containers for deployment → search "docker container kubernetes"
• Adjacent: security for production → search "ssl certificate security scan"

This produces 5–10 keywords the user never said but will need.

Step 3 — Check Local

Use your agent's native mechanism to see what skills are already installed. Examples for common agents:

# Claude Code
ls ~/.claude/skills/ 2>/dev/null && ls .claude/skills/ 2>/dev/null

# Codex / Cursor / Copilot (universal path)
ls ~/.agents/skills/ 2>/dev/null && ls .agents/skills/ 2>/dev/null

Or use the skills CLI (if installed):

npx skills check   # shows installed skills and available updates

Note any installed skills that match the user's domain/action.

Step 4 — Detect Available Search Tools

Before searching, quickly check which discovery tools are available in the user's environment. Don't assume any specific tool — detect and adapt:

# Check for the standard skills CLI
which npx 2>/dev/null && npx skills --version 2>/dev/null

# Check for other common skill managers
which openskills 2>/dev/null
npx load-skill --version 2>/dev/null

# Check local skill directories for discovery-oriented skills
ls ~/.claude/skills/ 2>/dev/null | grep -iE 'find|search|discover|browse'
ls ~/.agents/skills/ 2>/dev/null | grep -iE 'find|search|discover|browse'

Build a list of actually available search methods. Common ones include:

Tool	How to invoke	What it searches
npx skills find	npx skills find "<query>"	skills.sh registry
find-skills skill	Invoke the skill with keywords	GitHub SKILL.md files
openskills	openskills search "<query>"	Multiple registries
load-skill	npx load-skill search "<query>"	Broad skill index
Agent-native search	Varies by agent	Agent-specific

Step 5 — Search Remote

Search in priority order. find-skills first, then npx skills find, then any other tools detected in Step 4.

🥇 Primary — find-skills (if installed):

Start here. find-skills is the most recognized discovery tool and covers GitHub-indexed skills beyond the skills.sh registry. Run 3–5 searches with varied keywords.

🥈 Secondary — npx skills find (always available):

Fill gaps the primary search may have missed. Focus on keywords you haven't tried yet:

npx skills find "<exact keywords>"      # skills.sh registry
npx skills find "<broader keywords>"    # widen scope
npx skills find "<related domain>"      # adjacent areas

🥉 Fallback — openskills, load-skill, etc.:

Only if primary + secondary return sparse results and these tools are detected.

⚠️ Source attribution priority: if the same skill is found by multiple tools, report the highest-priority tool as the source (e.g. find-skills beats npx skills find). This gives users the most accessible discovery path.

Search strategy:

• Start specific, then broaden (e.g. "react deploy aws" → "deploy hosting" → "cloud infrastructure")
• Cover related domains (e.g. deploying → also search CI/CD, monitoring)
• Search in English even if the user writes in another language
• At least one broad category search to catch unexpected matches
• Deduplicate results before presenting (same skill found by multiple tools)

Step 6 — Rank & Present

Score results by relevance:

Tier	Criteria
🔥 Strong match	Directly addresses the user's stated task
👍 Worth considering	Covers a related or adjacent need
💡 Might be useful	Broader utility in the user's domain

For each recommendation, include:

• Skill name and install path (owner/repo@skill-name)
• One-line description
• Why it's relevant to this specific request
• The install command (npx skills add owner/repo@skill-name)
• Search source — the actual tool that found this skill (report whatever was used: npx skills find "...", find-skills, openskills, load-skill, etc. — multiple if found by more than one)

Step 7 — Ask the User

Present findings then ask:

• Which ones they want installed
• Whether to search in a different direction
• If they want more detail on any recommendation

---

Output Template

## 🔍 What I understood

**Task**: [one-line summary]
**Domain**: [tech area] | **Tools**: [tools] | **Action**: [action]
**Search keywords**: [list of keywords used]
**Search methods**: [list actually-used tools, e.g. `npx skills find`, `find-skills`, `openskills`]

## 📦 Already installed

| Skill | Useful because |
|-------|---------------|
| name | reason (or "No matching skills found") |

## 🌐 Recommendations

### 🔥 Strong matches
| Skill | Description | Why this helps | Source | Install |
|-------|------------|---------------|--------|---------|
| ... | ... | ... | `find-skills → "..."` | `npx skills add ...` |

### 👍 Worth considering
| Skill | Description | Why this helps | Source | Install |
|-------|------------|---------------|--------|---------|
| ... | ... | ... | `npx skills find → "..."` | `npx skills add ...` |

### 💡 Might be useful
| Skill | Description | Why this helps | Source | Install |
|-------|------------|---------------|--------|---------|
| ... | ... | ... | `openskills → "..."` | `npx skills add ...` |

## 💭 Recommendation

[1–2 sentence summary of the best install combo]

---
Want me to install any of these? Or search in a different direction?

---

Principles

1. Search generously, recommend selectively — Cast a wide net with keywords but only show the best matches.
2. Explain the "why" — Every recommendation must connect back to the user's request.
3. Think upstream and downstream — What comes before and after the user's stated task?
4. Bilingual coverage — Always try English keywords (the ecosystem is English-dominant).
5. Max 5 per tier — Keep recommendations focused; more isn't better.
6. Default to project install — Install skills into the project scope (no -g flag) so each project keeps its own dependencies. Users who prefer global can add -g.

---

Compatibility

Built for the Agent Skills open standard. Tested with:

• Claude Code — full support (symlink + universal install)
• Codex — universal install path
• Cursor — universal install path
• GitHub Copilot — universal install path
• Gemini CLI — requires skills CLI available
• Other agents supporting the Agent Skills standard

Requires: npx skills CLI (installed automatically on first use, or via npm install -g skills-cli).