SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsdiscover-skills

discover-skills

General & other

View repository

Updated 10-16 06:50

About this skill

Skill: Discovery Skill

Purpose

Help agents identify missing skills in a task and recommend specific installation steps. This skill discovers candidates and suggests what to install, but does not automatically install or inject skills.

Core Objective

Primary Objective: Provide agents with the top 1-3 skill recommendations and accurate installation commands to fill capability gaps.

Success Criteria(all requirements must be met):

  1. ✅ Perform discovery: Search local skills/INDEX.md and manifest.json, or search external directories as requested
  2. ✅ Best matches identified: Select 1-3 skills that match the task requirements based on name, description, and tags
  3. ✅ Recommendation explanation: Provides the reason why each skill matches the current task
  4. ✅ Installation commands provided: Includes the exact installation command for each recommended skill
  5. ✅ No automatic installation: Do not execute installation commands without explicit user confirmation

AcceptanceTest: Can the agent or user install the recommended skills using the provided commands without additional research?

Scope Boundaries (Scope Boundaries)

This skill is responsible for：

  • Discovering skills from local indexes or external directories
  • Matching task requirements with skill capabilities
  • Recommending the top 1-3 skill combinations with reasons
  • Providing accurate installation commands

This skill is not responsible for：

  • Automatically installing skills (usinginstall-rulesor manual installation)
  • Curating or reviewing existing skills (use "Curate Skills")
  • Refining or designing skills (use "Refine-Skill-Design")
  • Injecting skill content into the agent context (handled by the agent runtime)

Handoff point: When installation commands are provided as suggestions, hand them to the user for an installation decision or to "install-rules" for automatic installation (requires user confirmation).

Use Cases (Use Cases)

  • Initial bootstrapping: The agent starts with only this skill, then suggests which skills to install for the current task.
  • On-demand expansion: When a task requires unavailable skills, suggest the best matches and how to install them.
  • Capability discovery: Help users find relevant skills from local indexes or public directories.

Behavior (Behavior)

  1. Discovery: Prefer using local skills/INDEX.md and manifest.json as the capability map. If the user requests external options and network access is available, search public directories (e.g., SkillsMP).
  2. Matching: Match the user task against "name", "description", and "tags"; select the best 1-3 skills.
  3. Recommendation: Explain why each skill matches and whether it is local or external.
  4. Installation guidance: Provide the exact installation command for each recommended skill (e.g., "npx Skills add Owner/repo --skill name").
  5. Confirmation: If the user asks the agent to install, request explicit confirmation before running any command.

Input & Output

  • Input：
    • Description of the current task.
    • Optional: List of installed skills.
    • Optional: Allowed sources (local and public directories only).
  • Output：
    • Recommended skills with reasons.
    • Install command for each suggestion.

Constraints

Hard Boundaries

  • No automatic installation: Do not execute installation commands without explicit user confirmation.
  • No automatic injection: Do not automatically fetch or inject remote SKILL.md content.
  • No bulk discovery: Avoid listing large directories; return only the top 1-3 matches.

Skill Boundaries (Avoid overlap)

Do not do these (other skills can handle them)：

  • Install skills: Execute installation commands or modify the skill directory → useinstall-rulesor manual installation after user confirmation
  • Curate skills: Review skill quality, detect overlap, or score skills → use "Curate Skills"
  • Refine skills: Design, restructure, or improve existing skill content → use "Refine Skill Design"
  • Inject content: Load skill content into the Agent context or runtime → handled by the Agent runtime system

When to stop and hand off：

  • User says "install it" or "add that skill" → provide the installation command, request confirmation, hand off to "Install Rules" or manual installation
  • User asks "How do I improve this skill?" → hand off to "Refine Skill Design"
  • User asks "Is my skill quality good?" → hand off to "Curate Skills"
  • Provide suggestions along with installation commands → discovery complete, waiting for the user's decision

Self-Check

Core success criteria (all criteria must be met)

  • Perform discovery: Search local "skills/INDEX.md" and "manifest.json", or search external directories as required
  • Best match identified: Select 1-3 skills that match the task requirements by name, description, and tags
  • Suggestion explanation: Provided the reason why each skill matches the current task
  • Installation commands provided: Include the exact installation command for each recommended skill
  • No automatic installation: Do not execute installation commands without explicit user confirmation

Process quality check

  • Relevance: Are the suggestions closely related to the current task?
  • Actionable: Are the installation commands specific and correct?
  • Consent: Does the agent avoid running installations without explicit confirmation?
  • Source priority: Do I prefer the local index before searching external directories?
  • Conciseness: Do I limit recommendations to the top 1-3 matches, avoiding bulk directory listings?

Acceptance test

Can the agent or user install the recommended skills using the provided commands without additional research?

If no: the recommendation is incomplete. Verify that the installation command is correct and includes all necessary parameters.

If yes: discovery is complete. Wait for the user to decide on installation.

Examples (Examples)

Example 1: Recommend a local skill

  • Scenario: The user requests a standardized README file.
  • Steps：
    1. The agent checks skills/INDEX.md。
    2. and finds "generate-standard-readme".
    3. Recommend and provide:npx skills add nesnilnehc/ai-cortex --skill generate-standard-readme。

Example 2: Edge case - no local match

  • Scenario: The task is a niche task, and no local skill matches.
  • Expected: Provide 1-3 external suggestions via installation commands (if allowed), or say "no match found" and ask for clarification. Do not install automatically.

Appendix: Output Contract

When the skill produces recommendations, it follows this contract:

Element Requirement
Count Only the top 1-3 matches; no bulk catalog listing.
Per skill Name, rationale (why it matches), installation command (e.g. "npx Skills add Owner/repo --skill name").
Source Prefer local skills/INDEX.md and manifest.json; external only when the user requests it and the network is available.
Interaction Do not run installation commands without the user's explicit confirmation.