Skills Locator & Navigation

Use this skill to reliably find the right skill package inside a large skills corpus.

When this skill should trigger

Trigger whenever the user asks to:

• find a skill by topic, tool, framework, OS, or attack technique
• browse or navigate the skills tree
• identify where a specific tutorial/README skill landed
• shortlist candidate skills for a request
• inspect missing coverage or duplicates across many skills

Inputs

• A query, keyword list, or task description
• Optional scope path (subtree under the skills root)

Outputs

• Ranked list of matching skill directories
• For each match: path to

  SKILL.MD

  , short reason, and confidence (

  high|medium|low

  )
• If ambiguous: ask one clarification question and provide best-effort shortlist anyway

Workflow

1. Resolve skills root.
2. Run

   scripts/find_skill.py

   with the user query.
3. If needed, run

   scripts/list_skills.py

   for macro view (coverage by top-level area).
4. Return top 3-10 matches with exact paths.
5. If user picks one, open that skill and proceed with task execution.

Commands

Find by query

python3 scripts/find_skill.py --query "golden ticket kerberos" --top 8

Limit to subtree

python3 scripts/find_skill.py --query "token escalation" --scope "windows-hardening" --top 10

Show corpus overview

python3 scripts/list_skills.py --max-topics 40

Ranking behavior

find_skill.py

scores candidates using:

• exact/near token matches in path
• token matches in YAML

  name

  and

  description

• token matches in SKILL body

Path and metadata are weighted higher than generic body matches.

Response format for users

Use this compact format:

1. <path-to-skill-dir>

   — why it matches

2. <path-to-skill-dir>

   — why it matches

3. <path-to-skill-dir>

   — why it matches

Then ask: “Do you want me to open #1 or compare #1 vs #2 first?”

Notes

• Treat

  SKILL.MD

  and

  SKILL.md

  as equivalent.
• Prefer deterministic script output over ad-hoc grep when possible.
• If no strong match exists, return nearest matches and explicitly say coverage might be missing.