mcpField
S
skill-discovery

skill-discovery

github.com/matrixyAgent workflows2858 installsUpdated 21w ago
Repository

About skill-discovery

skill-discovery helps agents find reusable workflows for multi-step or repeatable tasks. It searches for skills based on the task description, presents matching results, and loads a skill only after the user approves.

Key features

  1. 01Searches skills from task descriptions
  2. 02Shows matches and confidence
  3. 03Explains how a skill runs
  4. 04Loads skills after user approval

Use cases

  1. 01Find an existing workflow for a multi-step task
  2. 02Discover reusable skills for recurring work
  3. 03Load a matching skill with user approval

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/matrixy/auto-skill --skill skill-discovery

After installing, type/skill-discoveryin Claude Code to confirm it loaded.

Details

From the project README

Auto-Skill Discovery

When the user requests a task that might match a previously learned workflow, search for relevant skills and offer to load them.

When to Invoke This Skill

Use this skill when the user's request:

  • Involves multiple steps (search + edit, read + modify, find + fix)
  • Sounds like a repeatable workflow
  • Mentions patterns like "the usual way" or "like before"
  • Could benefit from a structured approach

Do NOT use for:

  • Simple single-step tasks
  • Questions or explanations
  • Tasks the user wants done a specific way

Instructions

Step 1: Discover matching skills

Run the discovery script with the user's intent:

python scripts/discover_skill.py "<2-4 word summary of user's task>"

This outputs a formatted prompt showing:

  • Best matching skill name and confidence
  • What the skill does
  • How it runs (inline vs isolated)
  • A question asking if user wants to load it
Step 2: Present to user

Show the discovery output to the user. Wait for their response.

Step 3: If user approves, load the skill
python scripts/discover_skill.py "<same query>" --auto-load

This outputs the full skill content with clear delimiters. Display it directly - the formatted output contains the instructions you should follow.

Step 4: Follow the loaded skill

The skill content between the ====== delimiters contains your instructions. Follow them to complete the user's task.

Step 5: If user declines or no match

Proceed with the task normally using your standard approach.

Example Flow

User: "Find all the TODO comments and update them"

You: [Run discover_skill.py "find update todos"]

Output:
  ## Skill Discovery
  I found an auto-generated skill that matches your task:
  **search-and-fix-workflow** (confidence: 85%)
  > Search for issues in codebase and fix them systematically
  Would you like me to load this skill?

User: "Yes"

You: [Run discover_skill.py "find update todos" --auto-load]

Output:
  ======================================
  SKILL LOADED: search-and-fix-workflow
  ...instructions...
  ======================================

You: [Follow those instructions to complete the task]

Comments

Markdown supported