Hybrid Keyword + TF-IDF scoring ranks skills by relevance to conversation context.
- Ranked top-K shown verbosely in system prompt, rest as compact index
- Selection metadata (score, rank, total) embedded in skill tool output
- UI renders collapsible card with rank/score details
- triggers field in SKILL.md frontmatter boosts relevance
- maxShown config to control verbosity
- Tests for ranking, scoring, and caching

Normalize bun.lock after bun install — removes stale/redundant sha512
hashes for @solidjs/start (pkg.pr.new preview) and ghostty-web
(GitHub-hosted) entries that bun no longer writes for these sources.

Co-Authored-By: Oz <[email protected]>

Hybrid Keyword + TF-IDF scoring ranks skills by relevance to conversation context.
- Ranked top-K shown verbosely in system prompt, rest as compact index
- Selection metadata (score, rank, total) embedded in skill tool output
- UI renders collapsible card with rank/score details
- triggers field in SKILL.md frontmatter boosts relevance
- maxShown config to control verbosity
- Tests for ranking, scoring, and caching

…oved

- session/system.ts: update Interface.environment return type to Effect.Effect<string[]>
- session/prompt.ts: remove Effect.sync wrapper around sys.environment(model)
- skill/index.ts: remove unused ctx param from Skill.state function

Co-Authored-By: Oz <[email protected]>

…ements

Resolved one conflict in packages/opencode/src/skill/index.ts:
- Combined upstream's description-optional + built-in customize-opencode skill
  with local lastRankResult state, triggers parsing, rank/rankDetails/fmt(topK) APIs

All other files (prompt.ts, system.ts, config/skills.ts, tool/registry.ts,
message-part.tsx, test files) merged cleanly, preserving both upstream
improvements (scout tools, image handling, HTTP API hardening, websearch
provider label) and local skill-selection ranking enhancements.

Fixed: skillDocument/keywordScore now handle optional description safely.