Smart Skill Discovery

[Jeffallan]

Smart Skill Discovery — Feasibility Discussion

This issue consolidates the scope from #70 (context analysis algorithms), #71 (description indexing), and #72 (routing table intelligence), which have been closed as duplicates.

Problem

As the skill library grows, finding the right skill for a given task relies on static trigger keywords and description matching. This works for direct invocations but misses opportunities for contextual recommendations; suggesting skills the user didn't explicitly ask for but would benefit from.

Consolidated Scope

The following capabilities were originally spread across four issues:

• Semantic search/ranking — Vector embeddings for skill descriptions and reference content, enabling similarity-based matching beyond keyword triggers
• Context analysis — Analyze the current task, codebase, and conversation to infer which skills are relevant
• Description indexing — Keyword optimization and semantic indexing for faster, more accurate skill lookup
• Usage-history-based routing — Learn from past skill invocations to improve future suggestions

Open Questions

1. Tooling: What third-party tools/services would this require? Vector database (Pinecone, Chroma, etc.)? Embedding API? MCP tool for search?
2. Feasibility within Agent Skills spec: Can this be implemented as a skill/reference file, or does it require infrastructure outside the spec?
3. Scope reduction: Which subset of these capabilities delivers the most value with the least tooling dependency? Is keyword optimization (Skill metadata enhancement #69) sufficient, or is semantic search meaningfully better?
4. Integration point: Should this be a standalone discovery tool, or embedded into the existing skill routing/description system?

Context

• Skill metadata enhancement #69 (Skill metadata enhancement) covers structured metadata field improvements — achievable as pure skill work, in V0.7.0
• Cross-Domain Recommendations #65 (Cross-domain recommendations) and Enhanced Routing Logic #66 (Enhanced routing logic) are also in V0.7.0 as content work
• This issue covers the capabilities that may need third-party tooling beyond what skills alone can provide

Related Issues

• Skill metadata enhancement #69 — Skill metadata enhancement (V0.7.0)
• Cross-Domain Recommendations #65 — Cross-domain recommendations (V0.7.0)
• Enhanced Routing Logic #66 — Enhanced routing logic (V0.7.0)
• Improved context analysis algorithms #70, Better skill description indexing #71, Enhanced routing table intelligence #72 — Closed, scope merged here

[Jeffallan]

Updated per v0.5.0 roadmap consolidation. This issue now consolidates #70 (context analysis), #71 (description indexing), and #72 (routing intelligence). Full scope: semantic search/ranking (vector embeddings + MCP tool), context analysis, description indexing, usage-history-based routing. This requires third-party tooling exploration. Consider converting to a GitHub Discussion.