chore: update agent models and remove deprecated tools

- Changed agent models from anthropic to openai/gpt-5.2-codex for improved performance.
- Removed the deprecated beads agent file.
- Updated AGENTS.md to reflect changes in custom tools and deprecated commands.
- Added new swarmmail release functions for better file reservation management in the swarm plugin.

Author: joelhooks

AGENTS.md

@@ -197,8 +197,6 @@ The `opencode-swarm-plugin` provides type-safe, context-preserving wrappers. Alw
 
 ### Other Custom Tools
 
-- **cass_search, cass_view, cass_expand** - Search past agent sessions (use `hivemind_find` instead for unified interface)
-
 - **swarm_review, swarm_review_feedback** - Coordinator reviews worker output (3-strike rule)
 
 - **typecheck** - TypeScript check with grouped errors
@@ -208,34 +206,11 @@ The `opencode-swarm-plugin` provides type-safe, context-preserving wrappers. Alw
 - **repo-crawl\_\*** - GitHub API repo exploration
 - **repo-autopsy\_\*** - Clone & deep analyze repos locally
 - **pdf-brain\_\*** - PDF & Markdown knowledge base (supports URLs, `--expand` for context)
-- **ubs\_\*** - Multi-language bug scanner
-
-### DEPRECATED - Do Not Use Directly
-
-- ~~`bd` CLI commands~~ → Use `hive_*` plugin tools
-- ~~`bd-quick_*` tools~~ → Use `hive_*` plugin tools
-- ~~`beads_*` tools~~ → Use `hive_*` plugin tools (aliases deprecated)
-- ~~Agent Mail MCP tools~~ → Use `agentmail_*` plugin tools
-
-**Why?** Plugin tools have:
-
-- Type-safe Zod validation
-- Context preservation (hard caps on inbox, auto-release)
-- Learning integration (outcome tracking, pattern maturity)
-- UBS bug scanning on completion
-- Hivemind history queries for decomposition
   </tool_preferences>
 
 <context_preservation>
 **CRITICAL: These rules prevent context exhaustion. Violating them burns tokens and kills sessions.**
 
-### Agent Mail - MANDATORY constraints
-
-- **PREFER** `agentmail_inbox` plugin tool - enforces limit=5 and include_bodies=false automatically (plugin guardrails)
-- **ALWAYS** use `agentmail_summarize_thread` instead of fetching all messages in a thread
-- **ALWAYS** use `agentmail_read_message` for individual message bodies when needed
-- If using MCP tools directly: `include_bodies: false`, `inbox_limit: 5` max, `summarize_thread` over fetch all
-
 ### Documentation Tools (context7, effect-docs) - MANDATORY constraints
 
 - **NEVER** call these directly in the main conversation - they dump entire doc pages
@@ -434,43 +409,7 @@ git status   # MUST show "up to date with origin"
 hive_ready()
 ```
 
-## Agent Mail (via Plugin)
-
-<agent*mail_context>
-Agent Mail coordinates multiple agents working the same repo. \*\*Always use `agentmail*\*` plugin tools\*\* - they enforce context-safe limits (max 5 messages, no bodies by default).
-</agent_mail_context>
-
-### When to Use
-
-- Multiple agents working same codebase
-- Need to reserve files before editing
-- Async communication between agents
-
-### Workflow
-
-```
-# 1. Initialize (once per session)
-agentmail_init(project_path="/abs/path/to/repo", task_description="Working on X")
-# Returns: { agent_name: "BlueLake", project_key: "..." }
-
-# 2. Reserve files before editing
-agentmail_reserve(paths=["src/auth/**"], reason="bd-123: Auth refactor", ttl_seconds=3600)
 
-# 3. Check inbox (headers only, max 5)
-agentmail_inbox()
-
-# 4. Read specific message if needed
-agentmail_read_message(message_id=123)
-
-# 5. Summarize thread (PREFERRED over fetching all)
-agentmail_summarize_thread(thread_id="bd-123")
-
-# 6. Send message
-agentmail_send(to=["OtherAgent"], subject="Status", body="Done with auth", thread_id="bd-123")
-
-# 7. Release when done (or let swarm_complete handle it)
-agentmail_release()
-```
 
 ### Integration with Hive