Add intent-based smart skill search

[luongnv89]

Type

feature (high confidence)

Description

Add a smart search capability where users describe what they want to do in natural language, and ASM returns a ranked list of relevant skills — both from the index (installable) and from locally installed skills.

Current behavior: asm search matches against skill names and descriptions using keyword matching. Users need to know the right keywords to find relevant skills.

Desired behavior: Users describe their intent (e.g., "I want to review my code before merging") and ASM returns skills ranked by relevance to that intent, clearly indicating which are already installed and which can be installed. The user can then install and use the skill directly from the search results.

User Flow

1. Search by intent — User describes what they want to accomplish

   asm search "I want to review my code before merging"
   

2. Ranked results — ASM returns skills sorted by relevance, with install status

   1. ✓ code-review (installed)     — "Perform code reviews following best practices..."
   2.   pr-review (installable)     — "Review pull requests for bugs and quality..."
   3. ✓ simplify (installed)        — "Review changed code for reuse and quality..."
   

3. Install — User installs a skill directly from results

   asm install pr-review
   

4. Use — User invokes the skill in their AI tool (e.g., /pr-review in Claude Code)

Ranking Factors

• Semantic similarity between user intent and skill description/trigger conditions
• Skill category relevance (using the 22-category taxonomy from Adopt new 22-category taxonomy in catalog website and CLI search output #100)
• Popularity/verified status (from [FEATURE] Add verified/popular badges with a single skills.db and asm update-db command #21)
• Already-installed skills may be boosted or flagged for convenience

Reporter Context

feature request: smart skill search

• input: user describe what they want to do
• output: list of skills that they can install/or already have), ranking by relevant
• user install skill
• user use skill in their ai tool

Acceptance Criteria

• asm search accepts natural language intent descriptions as input (high confidence)
• Search results are ranked by relevance to the user's described intent (high confidence)
• Each result clearly indicates install status: installed vs. installable (high confidence)
• Users can install a skill directly from search results (e.g., asm install <name>) (high confidence)
• Ranking considers semantic similarity, not just keyword matching (medium confidence)

Metadata

• Priority: high
• Effort: L
• Related Issues: Adopt new 22-category taxonomy in catalog website and CLI search output #100 (22-category taxonomy for categorization), [FEATURE] Add verified/popular badges with a single skills.db and asm update-db command #21 (verified/popular badges for ranking signals), [FEATURE] Audit skill similarity to detect conflicting trigger descriptions #18 (skill similarity detection — related matching logic)
• Labels: feature, skill-discovery

Confidence Levels:

• Type: high confidence (explicit "feature request" with clear new capability)
• Acceptance criteria: high confidence (directly derived from the described user flow)