v1.0: Agent-to-agent protocol, skill manifests

[v1.0] - 2026-04-05

Added

• Weave inter-skill protocol spec (docs/protocol.md) — advertisement, request/response wire format, JSON schemas for WeaveAdvertisement, WeaveRequest, WeaveResponse
• weave.skill.json manifest support in all four adapters (Claude Code, Cursor, Codex, Windsurf) — name, version, capabilities, trigger_patterns, author, dependencies override adapter-inferred values
• Skill-specific manifest lookup (<stem>.skill.json preferred over weave.skill.json) for directories containing multiple skill files
• Dependency resolution at skill load time: SkillRegistry.get_by_name() and resolve_dependencies() — declared dependencies looked up by name, missing deps log a WARNING; weave load prints resolved link count
• Conflict detection in WeaveComposer: detect_conflicts() flags skill pairs with embedding similarity ≥ 0.9 AND opposing keyword pairs (always/never, use/avoid, enable/disable, etc.); WARNING logged per conflict, higher-score skill preferred automatically via score-sort
• 119 tests passing (1 skipped — ChromaDB integration, requires optional dep)

[Unreleased]