docs(reactions): document AI reaction agent (extra.mcp, reasoning, ai reply)

[ABazyk]

What

Syncs the plugin's reaction surface with the platform's AI reaction agent — the flow where a reaction flagged extra.mcp is processed by a Claude agent (hosted by claude-code-api) that acts on the platform via this MCP server, under the requesting user's access, and replies as a child ai reaction streaming its progress.

Changes

• mcp-server/internal/tools/reactions.go
  • extra object now documents mcp:boolean (true ⇒ hand the reaction to the AI agent).
  • createReaction / updateReaction gain a reasoning body param — {inProgress, thoughts:[{id,text,createdAt}]}, the agent's streamed progress/answer trace.
  • Notes that ai is the agent's reply type (usually platform-produced).
  • Descriptions/params only — no new operations. Drift gate (paths/methods) unaffected; make build && make vet and the drift test pass.
• skills/simulator-reactions/SKILL.md — new "AI agent reactions (extra.mcp)" section: how to invoke the agent, the ai reply, reasoning streaming (read mid-stream caveat), and a don't-loop note. Frontmatter untouched (no make discovery needed).
• docs/entities/reactions.md — extra/reasoning fields in the tree-edge table, the ai type, and a new "AI Reaction Agent (MCP)" section with the interaction-scheme diagram + field table.
• CHANGELOG.md — [Unreleased] entry.

Notes

• Documentation/contract sync only; the agent's runtime lives in pong-server (orchestrator) + claude-code-api and is tracked separately.
• No version bump: the manifests are currently inconsistent (.claude-plugin 1.7.1 vs .codex-plugin 1.10.0) — left untouched to avoid entangling this doc PR with that reconciliation.

🤖 Generated with Claude Code