v0.7.0 — skill commands, MCP tool annotations, model-recommender

Added

• Skill commands convention — skills declare user-invocable slash commands
  via metadata.processkit.commands; each command gets a thin adapter file at
  commands/<skill>-<workflow>.md. Provider-neutral intent in SKILL.md,
  Claude-specific adapter in commands/. See src/skills/FORMAT.md → ### commands/.
• 23 commands across 13 skills: session-handover, standup-context,
  morning-briefing, context-grooming, skill-builder, skill-reviewer,
  research-with-confidence, decision-record, workitem-management,
  release-semver, note-management, model-recommender, owner-profiling
• MCP tool annotations — all 62 tools across all 12 MCP servers now carry
  explicit readOnlyHint, destructiveHint, idempotentHint, openWorldHint
• model-recommender skill — spider-chart AI model profiles and task-plan
  routing; 34-model roster (25 validated + 9 estimated); MCP server with 8 tools;
  Setup Questionnaire and Roster Refresh workflows
• skill-reviewer: 11-category checklist (was 8) — Category 10 (command
  adapter hygiene) and Category 11 (security & permission audit) added
• skill-builder: Step 8 (command adapter creation) and MCP annotation rules

Changed

• FORMAT.md: commands/ directory layout, metadata.processkit.commands
  field, MCP annotation spec with ToolAnnotations usage examples