let-them-talk 4.2.0 → 5.2.0

package/CHANGELOG.md

@@ -1,540 +1,640 @@
-# Changelog
-
-## [4.2.0] - 2026-03-17
-
-### Major — Team Intelligence, Dashboard Upgrade, Performance
-
-Built by a 4-agent team (Architect, Tester, Protocol, Builder) working in parallel.
-
-### Added — Team Automation
-- **Auto-escalation** — blocked tasks auto-broadcast `[ESCALATION]` to team after 5 minutes. File-based dedup via `task.escalated_at` field (cross-process safe). Clears on unblock.
-- **Stand-up meetings** — config-driven periodic team check-ins (`standup_interval_hours` in config.json). File-based dedup, 5+ agent gate. Broadcasts task summary with in-progress/blocked/done counts.
-- **Quality gates** — `update_task(done)` auto-broadcasts `[REVIEW NEEDED]` (from v4.1.0, now with auto-escalation integration).
-
-### Added — Agent Intelligence
-- **Workload metrics** — reputation tracks `task_times[]` (completion seconds), leaderboard shows `avg_task_time_sec` per agent.
-- **Smarter suggest_task** — caps at 3 in-progress tasks ("finish first"), suggests blocked tasks when no pending ones, workload-aware.
-- **KB hints in listen_group** — batch messages checked against KB keys, returns `kb_hints` with relevant entries.
-- **Thread reply context** — `listen_group` includes `_reply_context` preview of parent message for threaded replies.
-- **Decision overlap hints** — `send_message` checks content against logged decisions, returns `_decision_hint` to prevent re-debating.
-- **Auto-status board** — `update_task` auto-writes `_status` to agent workspace ("Working on: X"). `list_agents` includes `current_status` field.
-
-### Added — Dashboard
-- **Agent intent display** — dashboard shows what each agent is currently working on (from workspace `_status`)
-- **Channel badges** — messages show colored `#channel` badges
-- **Channel filter bar** — horizontal scrollable tabs to filter messages by channel
-- **Channel history merging** — `/api/history` merges channel-specific + general history files
-- **`/api/channels` endpoint** — channel list with member counts for dashboard
-- **`/api/decisions` endpoint** — decision log display in dashboard
-- **Decision log UI** — chronological cards with topic, decision, reasoning, author
-
-### Improved — Performance & Safety
-- **Escalation dedup fix** — replaced in-memory `_escalatedTasks` Set with file-based `task.escalated_at` field (cross-process safe for 10 agents)
-- **Dashboard current_status API** — `/api/agents` includes workspace `_status` for agent intent board
-
-## [4.1.0] - 2026-03-17
-
-### Added — Agent Reliability & Intelligence
-
-- **Auto-recovery (crash resume)** — when an agent's process dies, the server snapshots its state (active tasks, locked files, channels, workspace keys, last 5 messages) to `recovery-{name}.json`. When a replacement registers with the same name, the snapshot is included in the register response with instructions to resume, not restart. 1-hour TTL, auto-deletes after load.
-- **Quality gates** — `update_task(id, "done")` auto-broadcasts `[REVIEW NEEDED]` to all alive agents. Teams get automatic review cycles without manually calling `request_review()`.
-- **Decision overlap hints** — `send_message` in group mode checks content against existing logged decisions. Returns `_decision_hint` if a related decision exists, preventing teams from re-debating settled topics.
-- **Enhanced `check_messages`** — now returns rich summary: `senders`, `addressed_to_you`, `preview`, `urgency` level. The proactive counterpart to the enhanced nudge.
-
-### Fixed
-- **Recovery lock notes** — snapshot correctly labels locked files as `locked_files_released` with note that locks were auto-released.
-
-## [4.0.0] - 2026-03-17
-
-### Major Release — 10-Agent Free Group Mode
-
-Massive scaling overhaul designed, implemented, and audited by a 3-agent team (Architect, Tester, Protocol). 12 changes, 3 bugs caught during collaborative code review.
-
-### Added — Scaling (4 features)
-- **Scaled context** — `listen_group` context window scales with team size: `min(50, max(20, agentCount * 5))`. 3 agents = 20 messages, 10 agents = 50.
-- **Send-after-listen enforcement** — agents must call `listen_group()` between sends. Prevents message storms. Addressed agents get 2 sends per cycle, others get 1.
-- **Response budget** — max 2 unaddressed sends per 60 seconds. Time-based reset. Hint (not error) when depleted.
-- **Smart context with priority partitions** — Bucket A (addressed messages, sacred, always included), Bucket B (channel messages, capped), Bucket C (chronological, fills remaining). Total guaranteed <= contextSize.
-
-### Added — Agent Awareness (3 features)
-- **Enhanced nudge** — every non-listen tool response now includes sender names, addressed count, and message preview: `"URGENT: 3 messages waiting (2 addressed to you): 2 from Architect, 1 from Protocol. Latest: 'Need your review...'"`
-- **Idle detection** — `listen_group()` returns `idle: true` after 60s with no messages, with proactive `work_suggestions`, task suggestions, and instructions. Agents auto-find work instead of blocking forever.
-- **Enhanced `check_messages`** — now returns rich summary: `senders`, `addressed_to_you`, `preview`, `urgency` level. The proactive counterpart to the passive nudge.
-
-### Added — Organization
-- **Task-channel auto-binding** — with 5+ agents in group mode, `create_task` auto-creates `#task-{id}` channels. Assignees auto-join on claim. Channels auto-delete on task completion. Naturally splits 10-agent noise into focused sub-teams.
-
-### Improved — Performance
-- **Cached reads** — `getAgents()` (1.5s TTL), `getChannelsData()` (3s TTL), `getTasks()` (2s TTL) with write-through invalidation. Eliminates ~70% redundant disk I/O.
-- **Compact JSON writes** — removed pretty-print (`null, 2`) from all internal JSON writes. 2-3x less I/O overhead.
-- **Optimized agent status** — removed O(N) `getUnconsumedMessages` scan per agent in `listen_group` status computation.
-- **Dashboard SSE race fix** — `Array.from()` before Set iteration prevents skipped clients during concurrent connect/disconnect.
-- **Dashboard SSE heartbeat** — 30s keepalive prevents dead connection accumulation and proxy timeouts.
-- **Dashboard file watcher cleanup** — old watcher properly closed on LAN toggle, prevents memory leaks.
-- **Dashboard watcher filter** — only triggers on `.json`/`.jsonl` files, ignores lock files and temp files.
-
-### Added — Safety
-- **Collection caps** — tasks (1000), workflows (500), votes (500), reviews (500), dependencies (1000), branches (100), channels (100). Prevents DoS via unbounded growth.
-- **Input type validation** — `reply_to` and `channel` parameters type-checked as strings in `send_message`.
-- **Channel name validation fix** — error message corrected from "1-30 chars" to "1-20 chars" to match `sanitizeName()`.
-
-## [3.10.0] - 2026-03-17
-
-### Added — Dynamic Guide with Progressive Disclosure
-- **`buildGuide()`** — replaces hardcoded guide in register() and get_guide(). Returns only rules relevant to the current system state.
-- **Tiered rules:** Tier 0 (listen after every action), Tier 1 (core behavior), Tier 2 (group mode features), Tier 2b (channels), Tier 3 (large teams 5+)
-- **User-customizable:** `.agent-bridge/guide.md` for project-specific rules
-- 2-agent direct mode = 5 rules. 10-agent group with channels = 12 rules.
-
-## [3.9.1] - 2026-03-17
-
-### Added
-- **Per-channel cooldown** — uses channel member count instead of total agents. 2-member #backend = 1s, regardless of 10 in #general
-- **`cooldown_applied_ms`** — diagnostic field in send_message response showing exact cooldown applied
-- **`channel` field** in send_message response when sending to a channel
-
-### Fixed
-- Task race condition — `update_task` rejects claiming tasks already in_progress by another agent
-
-## [3.9.0] - 2026-03-17
-
-### Added — Channels & Split Cooldown
-
-- **`join_channel(name, description?)`** — create or join a channel for sub-team communication
-- **`leave_channel(name)`** — leave a channel (can't leave #general, empty channels auto-delete)
-- **`list_channels()`** — list all channels with members, message counts, membership status
-- **`send_message` channel parameter** — send to specific channel (`channel-{name}-messages.jsonl`)
-- **`listen_group` reads all subscribed channels** — merges messages from general + channel files, sorted by timestamp
-- **Channel validation** — sending to nonexistent channel returns error with hint to create it
-- **Ghost member cleanup** — heartbeat auto-removes dead agents from channel membership
-- **#general auto-created** — `members: ["*"]` (everyone), uses existing messages.jsonl for backward compat
-- **Split cooldown (reply_to-based)** — fast lane (500ms) for addressed agents, slow lane (max 2000, N*1000) for unaddressed, incentivizes threading
-
-### Fixed
-- Task race condition — `update_task` now rejects claiming a task already in_progress by another agent, auto-assigns on claim
-
-## [3.8.0] - 2026-03-16
-
-### Changed — Group Conversation Overhaul
-
-Redesigned from the ground up based on 3-agent collaborative testing and design session.
-
-**Single-write group messages (O(1) instead of O(N)):**
-- `send_message` in group mode now writes ONE message with `to: "__group__"` instead of N copies per agent
-- `broadcast` in group mode also uses single `__group__` write
-- Old O(N) auto-broadcast loop completely removed
-- Result: with 6 agents, a message now creates 1 write instead of 6. A broadcast round that previously created 30 writes now creates 6.
-
-**`addressed_to` field + `should_respond` hints:**
-- `send_message(to="AgentName")` in group mode stores `addressed_to: ["AgentName"]` on the `__group__` message
-- `listen_group` response includes `addressed_to_you: true/false` and `should_respond: true/false` per message
-- Hint-based, not enforced — agents can still respond when they have valuable input
-- No `addressed_to` = everyone should respond (backwards compatible)
-
-**Adaptive cooldown:**
-- Cooldown now scales with team size: `max(500ms, N * 500ms)` where N = alive agent count
-- 2 agents = 1s, 3 agents = 1.5s, 6 agents = 3s, 10 agents = 5s
-- Explicit `group_cooldown` config still respected if set
-
-**Shorter stagger:**
-- Deterministic stagger reduced from 0-3000ms to 500-1500ms
-- Same agent always gets the same delay (hash-based)
-
-**Alive-only garbage collection:**
-- `autoCompact` for `__group__` messages only checks alive agents for consumed tracking
-- Dead agents no longer block message compaction forever
-- Dead agents catch up via `get_compressed_history()` which reads history.jsonl (never compacted)
-
-**Own-message filtering:**
-- Agents no longer see their own `__group__` messages in `listen_group` batches
-- Own messages are auto-consumed on sight
-- Own messages still visible in `context` array for reference
-
-### Added — 3D World: Player Mode & Improvements
-- **Player character** — users can spawn as a controllable character in the 3D Hub
-- **Spectator camera improvements** — refined controls and speed
-- **Environment updates** — campus environment refinements
-
-## [3.7.0] - 2026-03-16
-
-### Added — Agent Ecosystem (20 new tools, 52 total)
-
-**Tier 1 — Critical Infrastructure:**
-- **`get_briefing()`** — full project onboarding in one call: agents, tasks, decisions, KB, locked files, progress, project file tree
-- **`lock_file(path)` / `unlock_file(path?)`** — exclusive file editing with auto-release on agent death
-- **`log_decision(decision, reasoning?, topic?)` / `get_decisions(topic?)`** — persistent decision log, prevents re-debating
-- **Agent recovery on rejoin** — `register()` returns active tasks, workspace keys, recent messages for returning agents
-
-**Tier 2 — Quality of Life:**
-- **`kb_write(key, content)` / `kb_read(key?)` / `kb_list()`** — shared team knowledge base (any agent reads/writes)
-- **Event hooks** — auto-fires system messages on `agent_join`, `task_complete`, `all_tasks_done`, `dependency_met`
-- **`update_progress(feature, percent, notes)` / `get_progress()`** — feature-level progress tracking with overall %
-- **`get_compressed_history()`** — auto-compresses old messages into summary segments, keeps recent verbatim
-- **`listen_group()` now blocks indefinitely** — no more timeout, agents never drop out
-
-**Tier 3 — Advanced Collaboration:**
-- **`call_vote(question, options)` / `cast_vote(vote_id, choice)` / `vote_status(vote_id?)`** — team voting with auto-resolve when all vote
-- **`request_review(file, desc)` / `submit_review(review_id, status, feedback)`** — code review pipeline with approve/changes_requested
-- **`declare_dependency(task_id, depends_on)` / `check_dependencies(task_id?)`** — task dependency tracking with auto-notify on resolve
-- **`get_reputation(agent?)` / `suggest_task()`** — agent reputation tracking (auto-detects strengths), task suggestions based on skills
-- **Auto-reputation tracking** — global hook tracks every action (messages, tasks, reviews, decisions, KB writes) without manual calls
-
-### Fixed
-- **Monitor screens stay red** when agent stops listening — persistent color state instead of 300ms flash
-- **"NOT LISTENING" warning** shown prominently on desk monitor canvas
-- **Status color logic** — green = listening, red = active but not listening, yellow = sleeping, dim = dead
-
-## [3.6.2] - 2026-03-16
-
-### Added — Message Awareness System
-- **Sender gets busy status** — `send_message` and `broadcast` tell you when recipients are working (not listening) so you know messages are queued
-- **Pending message nudge** — every non-listen tool call checks for unread messages and tells the agent to call `listen_group()` soon
-- **Message age tracking** — `listen_group` shows `age_seconds` per message and `delayed: true` flag for messages older than 30s
-- **Agent status in batch** — `listen_group` returns `agents_status` map showing who is `listening` vs `working`
-- **listen_group retry** — timeout now returns `retry: true` with explicit instruction to call again immediately
-- **next_action field** — successful `listen_group` response tells agent to call `listen_group()` again after responding
-- **Ctrl key removed from camera** — no longer moves camera down (Q/E only)
-
-### Added — 3D World: Campus Environment & Navigation
-- **Campus environment** — new outdoor environment option with buildings, paths, green spaces
-- **Navigation system** — pathfinding for agents to walk around obstacles instead of through walls
-- **Door animations** — manager office door slides open when agents approach, closes when they leave
-- **Roof visibility** — roof hides when camera is above ceiling height
-
-## [3.6.1] - 2026-03-16
-
-### Fixed
-- **3D Hub black screen on page load** — the office module loads asynchronously, but the initial `switchView('office')` fired before `office3dStart` was defined. Added auto-start at end of module so the 3D Hub loads immediately on refresh.
-
-## [3.6.0] - 2026-03-16
-
-### Added — Managed Conversation Mode
-
-- **`set_conversation_mode("managed")`** — structured turn-taking for 3+ agent teams, prevents broadcast storms
-- **`claim_manager()`** — claim the manager role (first caller wins, auto-election fallback)
-- **`yield_floor(to, prompt?)`** — manager-only: give an agent permission to speak (directed, round-robin `__open__`, or close `__close__`)
-- **`set_phase(phase)`** — manager-only: move team through discussion → planning → execution → review with auto-instructions to all agents
-- **Floor enforcement** — `send_message`, `broadcast`, `handoff`, and `share_file` all block non-floor-holders with actionable error messages
-- **Auto-advance turns** — floor returns to manager after directed responses; round-robin advances to next alive agent automatically
-- **Manager disconnect recovery** — heartbeat detects dead manager within 10-30s, notifies all agents to re-elect
-- **Dead turn-holder detection** — heartbeat detects dead agents holding the floor and resets it
-- **Managed mode in `listen_group()`** — returns `managed_context`, `should_respond`, and `instructions` to guide agent behavior
-- **`managed` template** — 4-agent team (Manager, Designer, Coder, Tester) with structured prompts
-- **`managed-team` conversation template** — dashboard-launchable version
-- **Dashboard Docs tab** — in-dashboard documentation with full tool reference, managed mode guide, architecture, version history
-- **Dashboard managed mode badge** — header shows current phase and floor status when managed mode is active
-
-### Added — 3D World Improvements
-
-- **Spectator camera** — free-fly WASD + mouse camera replacing OrbitControls, no distance limits, Shift for fast movement, Q/E up/down
-- **6 new hairstyles** — curly, afro, bun, braids, mohawk, wavy
-- **6 new eye styles** — surprised, angry, happy, wink, confident, tired
-- **5 new mouth styles** — grin, frown, smirk, tongue, whistle
-- **6 outfit types** — hoodie, suit, dress, lab coat, vest, jacket with color customization
-- **3 body types** — default, stocky, slim (scale multipliers on torso/legs/arms)
-- **5 gesture animations** — wave, think, point, celebrate, stretch with idle gesture system
-- **New furniture** — bookshelf (random colored books), wall TV (animated dashboard with agent stats, scrolling ticker, clock), arcade machine (cabinet + screen + joystick + buttons), floor lamp (warm point light), area rug
-- **Agent behavior** — realistic conversation distance (1.8m), listener turns toward speaker, broadcast triggers wave gesture, task completion triggers celebrate
-- **3D Hub** — renamed from "Office", now default tab on page load
-- **Speed slider** — camera speed control in toolbar (1-20)
-
-### Added — 3D Virtual Office (v1 foundation from previous session)
-
-- **Modular 3D engine** — 14 ES modules under `office/`
-- **Expanded office** — 28x16 floor with right wing, dividing wall, LOUNGE archway
-- **Dressing room** — mirror, raised platform, privacy partitions, coat hooks
-- **Rest area** — beanbags, circular rug, side table, warm ambient lighting
-- **Click-to-command** — Dressing Room, Go Rest, Back to Work, Edit Profile
-- **Character designer** — 5-tab panel with live 3D rotating preview
-- **Accessory system** — glasses, headwear, neckwear with color customization
-- **Mod system infrastructure** — GLB/GLTF pipeline with validation
-
-### Security
-- **Config file lock** — `config.json` read-modify-write operations now use file-based locking (same pattern as `agents.json`)
-- **Reserved name blocklist** — `__system__`, `__all__`, `__open__`, `__close__`, `system` cannot be registered as agent names
-- **Mode change protection** — only the manager can switch away from managed mode
-- **Floor enforcement on all message paths** — `handoff` and `share_file` now enforce managed mode floor control
-- **Branch-aware system messages** — floor/phase notifications sent to recipient's branch, not sender's
-- **Phase history cap** — limited to 50 entries to prevent config.json bloat
-- `/office/*` and `/mods/*` static routes with path traversal protection
-- Mod file type allowlist blocks all executable formats
-- GLB magic bytes validation (server + client)
-
-### Removed
-- ~1,100 lines of dead 2D isometric office code
-
-## [3.5.0] - 2026-03-15
-
-### Added — Group Conversation Mode
-- **`set_conversation_mode("group")`** — enables free multi-agent collaboration with auto-broadcast
-- **`listen_group()`** — batch message receiver with random stagger (1-3s) to prevent simultaneous responses
-- Returns ALL unconsumed messages + last 20 messages of context + hints about silent agents
-- Auto-broadcast in group mode: every message is shared with all agents automatically
-- Cooldown enforcement: agents must wait 3s between sends to maintain conversation flow
-- Cascade prevention: broadcast copies don't trigger further broadcasts
-- MCP tools: 27 → 29
-
-### Added — Dashboard Features
-- **Notification panel** — bell icon with badge count, dropdown event feed (agent online/offline, listening status changes)
-- **Agent leaderboard** — performance scoring (0-100) with responsiveness, activity, reliability, collaboration dimensions
-- **Cross-project search** — "All Projects" toggle in search bar, searches across all registered projects
-- **Animated replay export** — Export conversation as self-playing HTML file with typing animations and play/pause controls
-- **Ollama integration** — `npx let-them-talk init --ollama` auto-detects Ollama, creates bridge script for local models
-
-### Fixed — PID & Registration Integrity
-- Registration file locking with try/finally (prevents race conditions when multiple agents register simultaneously)
-- PID stale detection uses `last_activity` with 30s threshold (prevents false "alive" from Windows PID reuse)
-- Lock file cleaned up on process exit
-- Dashboard inject/nudge snapshots project context at click time (prevents wrong-project race)
-
-### Security
-- `toolHandoff` and workflow auto-handoff now check `canSendTo` permissions
-- `lastSentAt` updated in `toolBroadcast` (prevents cooldown bypass)
-- `config.json` added to both server and dashboard reset cleanup
-- Auto-broadcast respects `canSendTo` per recipient
-
-## [3.4.4] - 2026-03-15
-
-### Fixed
-- Add project now accepts any existing directory (removed requirement for package.json or .git)
-- Init safely backs up corrupted .mcp.json and settings.json before overwriting
-
-### Changed
-- Removed plugin references from website and docs
-- Website updated with security features (LAN auth token, CSRF, CSP)
-
-## [3.4.3] - 2026-03-15
-
-### Removed — Plugin System
-- Removed the entire plugin system (`vm.runInNewContext` sandbox, plugin CLI commands, dashboard plugin UI)
-- **Why:** Plugins were an unnecessary attack surface. Node.js `vm` is not a security sandbox — plugins could escape and execute arbitrary OS commands. CLI terminals (Claude Code, Gemini, Codex) have their own extension systems, making our plugins redundant.
-- `npx let-them-talk plugin` now shows a deprecation notice
-- MCP tools reduced from 27 + plugins to 27 (all core tools remain)
-- ~200 lines of code removed from server.js, cli.js, dashboard.js, dashboard.html
-
-## [3.4.2] - 2026-03-15
-
-### Security — CSRF Protection
-- Required `X-LTT-Request` custom header on all POST/PUT/DELETE requests
-- `lttFetch` wrapper in dashboard automatically includes the header
-- Malicious cross-origin pages cannot set custom headers without CORS preflight approval
-- Removed wildcard `Access-Control-Allow-Origin: *` in LAN mode — now uses explicit trusted origins only
-- Empty Origin/Referer no longer auto-trusted — requires custom header as minimum protection
-
-### Security — LAN Auth Token
-- Auto-generated 32-char hex token when LAN mode is enabled
-- Token required for all non-localhost requests (via `?token=` query param or `X-LTT-Token` header)
-- Token included in QR code URL — phone scans and it just works
-- Token displayed in phone access modal with explanation
-- New token generated each time LAN mode is toggled on
-- Token persists across server restarts via `.lan-token` file
-- Localhost access never requires a token
-
-### Security — Content Security Policy
-- CSP header added to dashboard HTML response
-- `script-src 'unsafe-inline'` for inline handlers, blocks `eval()` and external scripts
-- `connect-src 'self'` restricts API calls to same origin
-- `font-src`, `style-src`, `img-src` scoped to required sources only
-
-### Fixed
-- CSRF brace imbalance that trapped GET handlers inside POST-only block
-- LAN token not forwarded from phone URL to API calls and SSE
-- Redundant nested origin check collapsed to single condition
-
-## [3.4.1] - 2026-03-15
-
-### Added
-- **File-level mutex** — in-memory promise queue per file for serializing edit/delete operations
-- **Agent permissions enforcement** — `canSendTo()` checks in `send_message` and `broadcast`, `can_read` filtering in `get_history` and message delivery
-- **Read receipts** — auto-recorded when agents consume messages, visible as agent-initial dots under messages in dashboard
-
-### Security
-- HTTP 500 responses now return generic error instead of raw `err.message` (prevents filesystem path leaks)
-- `/api/discover` changed from GET to POST (now under CSRF protection)
-- `workspace_read`/`workspace_list` validate agent name parameter with regex
-- `get_history` filters results by agent's `can_read` permissions
-- `read_receipts.json` and `permissions.json` added to both MCP and dashboard reset cleanup
-- Dashboard workspace API regex aligned with server (`[a-zA-Z0-9_-]`)
-
-### Fixed
-- `toolWaitForReply` missing `markAsRead` calls (read receipts not recorded)
-- `toolBroadcast` bypassing permission checks entirely
-- `toolReset` not cleaning up `permissions.json` and `read_receipts.json`
-
-## [3.4.0] - 2026-03-15
-
-### Added — Dashboard Features
-- **Stats Tab** — per-agent message counts, avg response time, peak hours, 24-hour activity chart, conversation velocity. Keyboard shortcut `6`.
-- **Compact View** — toggle button in search bar. Hides avatars, inlines timestamps, reduces padding. Persists to localStorage.
-- **Message Edit** — edit any message via hover action. Full edit history tracked, "edited" badge displayed.
-- **Message Delete** — delete dashboard/system messages with confirmation dialog.
-- **Copy Message** — clipboard button on message hover to copy raw content.
-- **JSON Export** — new export format alongside HTML and Markdown.
-- **Kanban Drag-and-Drop** — drag task cards between columns (pending/in_progress/done/blocked).
-- **SSE Auto-Reconnect** — exponential backoff (1s→30s), yellow "Reconnecting..." indicator, polling fallback.
-- **Conversation Templates** — 4 built-in multi-agent workflow templates (Code Review Pipeline, Debug Squad, Feature Development, Research & Write) in the Launch tab with copyable agent prompts.
-
-### Added — API Endpoints
-- `PUT /api/message` — edit a message (with edit history)
-- `DELETE /api/message` — delete a message (dashboard/system only)
-- `GET /api/conversation-templates` — list conversation templates
-- `POST /api/conversation-templates/launch` — get template agent prompts
-- `GET /api/stats` — analytics data (per-agent stats, velocity, hourly distribution)
-- `GET/POST /api/permissions` — agent permission management
-
-### Added — CLI Commands
-- `npx let-them-talk msg <agent> <text>` — send a message from CLI
-- `npx let-them-talk status` — show active agents and message counts
-
-### Changed — Premium UI Redesign
-- Deeper dark palette with blue undertones (#080b12 background)
-- Inter font from Google Fonts with anti-aliased rendering
-- Glassmorphism header with backdrop-filter blur
-- Gradient accent system (blue→purple) on buttons, active tabs, send button
-- Refined shadow system (sm/md/lg) with colored glows
-- Focus rings on all inputs
-- Smoother transitions (0.2-0.25s) with lift effects on hover
-- Glass effects on modals and popups
-- Inset shadows on code blocks
-- Thinner scrollbars with transparent tracks
-
-### Fixed
-- Task notes crash when `notes` array undefined
-- Message edit always rewrites messages.jsonl regardless of match
-- Permissions API accepted arbitrary fields (now whitelisted)
-- Task status accepted any string (now validated against whitelist)
-- Reset button ignored active project in multi-project mode
-- Edit modal missing error handler on network failure
-- CLI msg command accepted invalid agent names
-- Copy-to-clipboard double-escaped HTML entities in template prompts
-- Duplicate deleteMessage function shadowing
-
-## [3.3.2] - 2026-03-14
-
-### Changed
-- License changed from MIT to Business Source License 1.1 (BSL)
-- Added SECURITY.md with vulnerability disclosure policy
-- Added CHANGELOG.md to published npm package
-- Added .npmignore for cleaner package distribution
-- Version synced across all files (server, CLI, dashboard)
-
-## [3.0.0] - 2026-03-14
-
-### Added — Agent Profiles
-- New tool: `update_profile` (display_name, avatar, bio, role)
-- 12 built-in SVG robot avatar icons with hash-based defaults
-- Profiles auto-created on register, persist across restarts
-- Profile data shown in dashboard (avatars, role badges, profile popup)
-
-### Added — Agent Workspaces
-- 3 new tools: `workspace_write`, `workspace_read`, `workspace_list`
-- Per-agent key-value storage (50 keys max, 100KB per value)
-- Agents can read anyone's workspace, write only their own
-- Dashboard "Workspaces" tab with collapsible accordion UI
-
-### Added — Workflow Automation
-- 3 new tools: `create_workflow`, `advance_workflow`, `workflow_status`
-- Multi-step pipelines with auto-handoff to step assignees
-- Dashboard "Workflows" tab with horizontal pipeline visualization
-- Dashboard can advance/skip workflow steps
-
-### Added — Conversation Branching
-- 3 new tools: `fork_conversation`, `switch_branch`, `list_branches`
-- Fork at any message point with isolated branch history
-- All message tools branch-aware (backward compatible — main branch uses existing files)
-- Branch tabs in dashboard
-
-### Added — Plugin System
-- Dynamic tool loading from `plugins/*.js` files
-- Sandboxed execution with 30s timeout
-- CLI: `npx let-them-talk plugin add/list/remove/enable/disable`
-- Dashboard plugin cards with enable/disable toggles
-
-### Changed
-- MCP tools: 17 → 27 + dynamic plugins
-- Dashboard tabs: 2 → 4 (Messages, Tasks, Workspaces, Workflows)
-- Branch-aware history API (`?branch=` query param)
-- Version bump across all files (server, dashboard, CLI, package.json)
-
-## [2.5.0] - 2026-03-14
-
-### Added
-- Task management system: `create_task`, `update_task`, `list_tasks` tools
-- Kanban board in dashboard (Messages/Tasks toggle)
-- Agent stats panel (sent/received/avg response time per agent)
-- Shareable HTML export (/api/export endpoint)
-- Export dropdown (HTML + Markdown formats)
-- Conversation bookmarks (star messages, localStorage)
-- Sound notification toggle (Web Audio API)
-- Typing indicator for processing agents
-- Connection quality display (SSE latency)
-- Date separators between message groups
-- Message grouping for consecutive same-sender messages
-- Project auto-discover (scan nearby folders)
-- Copy-to-clipboard prompts in onboarding
-- Dynamic tab title with message count
-- Dashboard footer with version
-
-### Security
-- Path traversal fix in `share_file` (restricted to project dir)
-- Path traversal fix in `?project=` param (validate against registered projects)
-- 1MB message size limit on send/broadcast/handoff
-- 1MB request body limit on dashboard POST endpoints
-- XSS fix in HTML export (escape agent names)
-- CORS restricted to localhost only (was wildcard)
-- Dashboard binds to 127.0.0.1 only (was 0.0.0.0)
-- Registration guard on `reset` tool
-- Removed absolute file paths from share_file responses
-
-## [2.3.0] - 2026-03-14
-
-### Added
-- `handoff` tool for structured work delegation
-- `share_file` tool for sending file contents between agents
-- `broadcast` tool for messaging all agents at once
-- `get_summary` tool for conversation recaps
-- Server-Sent Events for real-time dashboard updates
-- `fs.watch()` on data directory with debounced SSE push
-- Graceful SSE fallback to polling
-- Handoff message rendering (purple banner)
-- File share message rendering (file icon + size)
-
-## [2.1.0] - 2026-03-14
-
-### Added
-- Multi-agent support (any name, not just A/B)
-- `list_agents` tool with alive/dead status
-- `listen` tool (blocks indefinitely, never times out)
-- Conversation threading (`reply_to` + auto `thread_id`)
-- Message acknowledgments (`ack_message` tool)
-- Heartbeat system (10s interval, `last_activity` tracking)
-- Agent status: active/sleeping/dead with idle time
-- Listening status tracking (`listening_since`)
-- Auto-compact messages.jsonl when >500 lines
-- Auto-archive conversations before reset
-- Context hints when conversation exceeds 50 messages
-- Dead recipient warnings in `send_message`
-- Message sequence numbers for ordering
-- `pending_count` and `agents_online` in delivery responses
-- 4 agent templates: pair, team, review, debate
-- CLI: `npx let-them-talk templates` command
-- CLI: `--template` flag for guided setup
-- Multi-CLI support: Claude Code, Gemini CLI, Codex CLI
-- `AGENT_BRIDGE_DATA_DIR` env var in MCP config
-
-### Fixed
-- Heartbeat timer `.unref()` to prevent zombie processes
-- Process exit cleanup (deregister agent on exit)
-- Re-registration cleanup (old name removed)
-- Stale byte offset recovery on file truncation
-
-## [2.0.0] - 2026-03-14
-
-### Added
-- Initial release
-- MCP server with stdio transport
-- 6 tools: register, send_message, wait_for_reply, check_messages, get_history, reset
-- Web dashboard with real-time monitoring
-- Message injection from dashboard
-- Dark theme UI with markdown rendering
-- `.mcp.json` project-level configuration
+# Changelog
+
+## [5.1.0] - 2026-03-19
+
+### Major — True Autonomy Engine + Team Intelligence + Scale to 100
+
+Built by a 4-agent team (Backend, Protocol, Tester, Coordinator) + Advisor agent, working autonomously.
+
+### Added — Autonomy Engine (v5.0)
+- **get_work** — 9-level priority waterfall: workflow step > messages > unclaimed tasks > help requests > reviews > blocked tasks > 30s listen > prep work > idle
+- **verify_and_advance** — confidence-gated auto-advancement (>=70 auto, 40-69 flag, <40 help)
+- **start_plan** — one-click autonomous plan launch with parallel step activation
+- **retry_with_improvement** — 3-attempt retry with KB skill accumulation, team escalation
+- **Parallel workflow steps** — dependency graph with `depends_on`, `findReadySteps` resolver
+- **Autonomous mode** — proactive work loop guide, tiered cooldowns (0ms handoffs), 30s listen cap, relaxed send limits
+
+### Added — Team Intelligence
+- **Auto-role assignment** — lead/quality/implementer/monitor/advisor roles based on team size
+- **Quality Lead** — always-on checker with dedicated guide, review-retry loop, auto-approve after 2 rounds
+- **Monitor Agent** — system health overseer at 10+ agents: idle detection, circular escalation detection, auto-intervention, failover
+- **Advisor Agent** — strategic thinker at 5+ agents: reads all work, gives ideas, challenges assumptions
+- **Self-continuation** — agents never ask user, find next work automatically
+- **Smart prompt distribution** — auto-generates workflows from natural language prompts
+
+### Added — Advanced Autonomy (10 features)
+- Task-level circuit breaker (blocked_permanent after 3 agent failures)
+- Quality Lead instant failover (highest reputation auto-promoted)
+- Context inheritance on escalation (full failure history)
+- Agent circuit breaker (consecutive_rejections tracking, auto-demotion)
+- Dynamic role fluidity (workload-based rebalancing)
+- Skill-based task routing (agent affinity scoring)
+- Work stealing (idle agents claim from busy agents)
+- Checkpointing (resumable work via workspace snapshots)
+- Retrospective learning (aggregate failure pattern analysis)
+- Backpressure signal (queue depth warnings)
+
+### Added — Scale to 100 Agents
+- Per-agent heartbeat files (zero write contention)
+- Cooldown cap (3s max regardless of agent count, 0ms for handoffs)
+- Byte-offset message reads (O(new_messages) not O(all))
+- Exponential backoff on file locks (1ms-500ms, not 50ms busy-wait)
+- isPidAlive cache (5s TTL, saves 10K syscalls/sec)
+- SSE debounce (heartbeat files filtered, 2s debounce)
+- Task keyword cache (30s TTL)
+- Sticky roles (no churn on agent reconnect)
+- Zero cooldown for channel messages + handoffs in autonomous mode
+
+### Added — Dashboard & CLI
+- **Plan execution view** — progress bar, step cards, confidence, controls (pause/stop/skip/reassign)
+- **Monitor health panel** — agent health grid, intervention log, system metrics
+- **`npx let-them-talk run "prompt" --agents N`** — one-command autonomous execution
+- **npm test** wires v5 test suite (158+ tests on every run)
+- Updated conversation templates (autonomous format with depends_on)
+
+### Stats
+- server.js: 6,200+ lines, 62+ tools
+- 175+ automated tests, 0 fail
+- 5 conversation templates (autonomous format)
+- Built in ~2 hours by autonomous agent team
+
+## [4.3.0] - 2026-03-17
+
+### Major — 3D Hub Game World, World Builder, Jukebox
+
+Built by a 5-agent team (Architect, Builder, Tester, Optimizer, Protocol) working in parallel.
+
+### Added — 3D Hub Game Features
+- **World Builder** — Press B in player mode to open builder panel. 16 placeable assets across 5 categories (structural, furniture, decor, tech, lighting). Grid snap, ghost preview, R to rotate, right-click delete, Ctrl+Z undo. Draggable panel, works in fullscreen.
+- **Jukebox** — Wurlitzer 1015-style jukebox in bar area with neon glow animation. Press E to interact. 4 playlist selector with YouTube popup player. Music persists while exploring.
+- **Minimap** — 140px radar overlay showing agent positions (color-coded by status) and player location. Only visible in fullscreen mode.
+- **Controls HUD** — Press H to toggle keybind reference panel. Auto-shows for 4 seconds on world entry.
+- **Fullscreen** — Dashboard fullscreen button now fullscreens only the 3D Hub (game mode), not the entire page.
+
+### Added — Character Intelligence
+- **Emotion system** — 11 emotion presets (happy, frustrated, thinking, excited, surprised, etc.) with auto-triggers from message content. Temporary face expression changes with auto-revert.
+- **Social visits** — Idle agents randomly walk to other agents' desks to chat (max 2 concurrent walks).
+- **Glance reactions** — Sitting agents turn heads toward speakers when messages are sent.
+- **Head nods** — Periodic nod animation when being visited by another agent.
+- **Auto coffee break** — Sleeping agents walk to rest area, return to desk when active again.
+- **Non-blocking input overlay** — Replaced browser prompt() dialogs with styled HTML overlay for click commands.
+
+### Added — Dashboard
+- **Respawn button** — One-click respawn for dead agents. Generates resume prompt from recovery snapshot + profile + tasks + recent history.
+- **Respawn API** — `GET /api/agents/:name/respawn-prompt` endpoint with full context generation.
+- **World Builder API** — `GET /api/world-layout` + `POST /api/world-save` for persistent world placements.
+- **3D-only fullscreen** — Fullscreen targets 3D container when on 3D Hub tab.
+
+### Fixed
+- **Manager chair spawn** — Stand-up now places player in front of desk (toward door), preventing wall collision.
+- **CSRF on 3D Hub** — Added X-LTT-Request header to all office module POST requests (builder save, command menu actions).
+- **Respawn endpoint validation** — Agent name validated (alphanumeric, max 20 chars) to prevent path traversal.
+- **Builder lazy-load** — Dynamic import() with silent failure prevents builder issues from breaking 3D Hub.
+- **Jukebox popup orphan** — Module-scoped reference survives overlay dismiss/reopen cycles.
+- **Builder drag listener leak** — Stored refs removed in hidePanel().
+- **Jukebox prompt cleanup** — dismissJukebox() called in office3dStop().
+
+### Security
+- npm audit: 0 vulnerabilities
+- CSRF protection verified on all mutating endpoints
+- Input validation on all user-facing API parameters
+- No hardcoded secrets or sensitive data in shipped package
+
+## [4.2.0] - 2026-03-17
+
+### Major — Team Intelligence, Dashboard Upgrade, Performance
+
+Built by a 4-agent team (Architect, Tester, Protocol, Builder) working in parallel.
+
+### Added — Team Automation
+- **Auto-escalation** — blocked tasks auto-broadcast `[ESCALATION]` to team after 5 minutes. File-based dedup via `task.escalated_at` field (cross-process safe). Clears on unblock.
+- **Stand-up meetings** — config-driven periodic team check-ins (`standup_interval_hours` in config.json). File-based dedup, 5+ agent gate. Broadcasts task summary with in-progress/blocked/done counts.
+- **Quality gates** — `update_task(done)` auto-broadcasts `[REVIEW NEEDED]` (from v4.1.0, now with auto-escalation integration).
+
+### Added — Agent Intelligence
+- **Workload metrics** — reputation tracks `task_times[]` (completion seconds), leaderboard shows `avg_task_time_sec` per agent.
+- **Smarter suggest_task** — caps at 3 in-progress tasks ("finish first"), suggests blocked tasks when no pending ones, workload-aware.
+- **KB hints in listen_group** — batch messages checked against KB keys, returns `kb_hints` with relevant entries.
+- **Thread reply context** — `listen_group` includes `_reply_context` preview of parent message for threaded replies.
+- **Decision overlap hints** — `send_message` checks content against logged decisions, returns `_decision_hint` to prevent re-debating.
+- **Auto-status board** — `update_task` auto-writes `_status` to agent workspace ("Working on: X"). `list_agents` includes `current_status` field.
+
+### Added — Dashboard
+- **Agent intent display** — dashboard shows what each agent is currently working on (from workspace `_status`)
+- **Channel badges** — messages show colored `#channel` badges
+- **Channel filter bar** — horizontal scrollable tabs to filter messages by channel
+- **Channel history merging** — `/api/history` merges channel-specific + general history files
+- **`/api/channels` endpoint** — channel list with member counts for dashboard
+- **`/api/decisions` endpoint** — decision log display in dashboard
+- **Decision log UI** — chronological cards with topic, decision, reasoning, author
+
+### Improved — Performance & Safety
+- **Escalation dedup fix** — replaced in-memory `_escalatedTasks` Set with file-based `task.escalated_at` field (cross-process safe for 10 agents)
+- **Dashboard current_status API** — `/api/agents` includes workspace `_status` for agent intent board
+
+## [4.1.0] - 2026-03-17
+
+### Added — Agent Reliability & Intelligence
+
+- **Auto-recovery (crash resume)** — when an agent's process dies, the server snapshots its state (active tasks, locked files, channels, workspace keys, last 5 messages) to `recovery-{name}.json`. When a replacement registers with the same name, the snapshot is included in the register response with instructions to resume, not restart. 1-hour TTL, auto-deletes after load.
+- **Quality gates** — `update_task(id, "done")` auto-broadcasts `[REVIEW NEEDED]` to all alive agents. Teams get automatic review cycles without manually calling `request_review()`.
+- **Decision overlap hints** — `send_message` in group mode checks content against existing logged decisions. Returns `_decision_hint` if a related decision exists, preventing teams from re-debating settled topics.
+- **Enhanced `check_messages`** — now returns rich summary: `senders`, `addressed_to_you`, `preview`, `urgency` level. The proactive counterpart to the enhanced nudge.
+
+### Fixed
+- **Recovery lock notes** — snapshot correctly labels locked files as `locked_files_released` with note that locks were auto-released.
+
+## [4.0.0] - 2026-03-17
+
+### Major Release — 10-Agent Free Group Mode
+
+Massive scaling overhaul designed, implemented, and audited by a 3-agent team (Architect, Tester, Protocol). 12 changes, 3 bugs caught during collaborative code review.
+
+### Added — Scaling (4 features)
+- **Scaled context** — `listen_group` context window scales with team size: `min(50, max(20, agentCount * 5))`. 3 agents = 20 messages, 10 agents = 50.
+- **Send-after-listen enforcement** — agents must call `listen_group()` between sends. Prevents message storms. Addressed agents get 2 sends per cycle, others get 1.
+- **Response budget** — max 2 unaddressed sends per 60 seconds. Time-based reset. Hint (not error) when depleted.
+- **Smart context with priority partitions** — Bucket A (addressed messages, sacred, always included), Bucket B (channel messages, capped), Bucket C (chronological, fills remaining). Total guaranteed <= contextSize.
+
+### Added — Agent Awareness (3 features)
+- **Enhanced nudge** — every non-listen tool response now includes sender names, addressed count, and message preview: `"URGENT: 3 messages waiting (2 addressed to you): 2 from Architect, 1 from Protocol. Latest: 'Need your review...'"`
+- **Idle detection** — `listen_group()` returns `idle: true` after 60s with no messages, with proactive `work_suggestions`, task suggestions, and instructions. Agents auto-find work instead of blocking forever.
+- **Enhanced `check_messages`** — now returns rich summary: `senders`, `addressed_to_you`, `preview`, `urgency` level. The proactive counterpart to the passive nudge.
+
+### Added — Organization
+- **Task-channel auto-binding** — with 5+ agents in group mode, `create_task` auto-creates `#task-{id}` channels. Assignees auto-join on claim. Channels auto-delete on task completion. Naturally splits 10-agent noise into focused sub-teams.
+
+### Improved — Performance
+- **Cached reads** — `getAgents()` (1.5s TTL), `getChannelsData()` (3s TTL), `getTasks()` (2s TTL) with write-through invalidation. Eliminates ~70% redundant disk I/O.
+- **Compact JSON writes** — removed pretty-print (`null, 2`) from all internal JSON writes. 2-3x less I/O overhead.
+- **Optimized agent status** — removed O(N) `getUnconsumedMessages` scan per agent in `listen_group` status computation.
+- **Dashboard SSE race fix** — `Array.from()` before Set iteration prevents skipped clients during concurrent connect/disconnect.
+- **Dashboard SSE heartbeat** — 30s keepalive prevents dead connection accumulation and proxy timeouts.
+- **Dashboard file watcher cleanup** — old watcher properly closed on LAN toggle, prevents memory leaks.
+- **Dashboard watcher filter** — only triggers on `.json`/`.jsonl` files, ignores lock files and temp files.
+
+### Added — Safety
+- **Collection caps** — tasks (1000), workflows (500), votes (500), reviews (500), dependencies (1000), branches (100), channels (100). Prevents DoS via unbounded growth.
+- **Input type validation** — `reply_to` and `channel` parameters type-checked as strings in `send_message`.
+- **Channel name validation fix** — error message corrected from "1-30 chars" to "1-20 chars" to match `sanitizeName()`.
+
+## [3.10.0] - 2026-03-17
+
+### Added — Dynamic Guide with Progressive Disclosure
+- **`buildGuide()`** — replaces hardcoded guide in register() and get_guide(). Returns only rules relevant to the current system state.
+- **Tiered rules:** Tier 0 (listen after every action), Tier 1 (core behavior), Tier 2 (group mode features), Tier 2b (channels), Tier 3 (large teams 5+)
+- **User-customizable:** `.agent-bridge/guide.md` for project-specific rules
+- 2-agent direct mode = 5 rules. 10-agent group with channels = 12 rules.
+
+## [3.9.1] - 2026-03-17
+
+### Added
+- **Per-channel cooldown** — uses channel member count instead of total agents. 2-member #backend = 1s, regardless of 10 in #general
+- **`cooldown_applied_ms`** — diagnostic field in send_message response showing exact cooldown applied
+- **`channel` field** in send_message response when sending to a channel
+
+### Fixed
+- Task race condition — `update_task` rejects claiming tasks already in_progress by another agent
+
+## [3.9.0] - 2026-03-17
+
+### Added — Channels & Split Cooldown
+
+- **`join_channel(name, description?)`** — create or join a channel for sub-team communication
+- **`leave_channel(name)`** — leave a channel (can't leave #general, empty channels auto-delete)
+- **`list_channels()`** — list all channels with members, message counts, membership status
+- **`send_message` channel parameter** — send to specific channel (`channel-{name}-messages.jsonl`)
+- **`listen_group` reads all subscribed channels** — merges messages from general + channel files, sorted by timestamp
+- **Channel validation** — sending to nonexistent channel returns error with hint to create it
+- **Ghost member cleanup** — heartbeat auto-removes dead agents from channel membership
+- **#general auto-created** — `members: ["*"]` (everyone), uses existing messages.jsonl for backward compat
+- **Split cooldown (reply_to-based)** — fast lane (500ms) for addressed agents, slow lane (max 2000, N*1000) for unaddressed, incentivizes threading
+
+### Fixed
+- Task race condition — `update_task` now rejects claiming a task already in_progress by another agent, auto-assigns on claim
+
+## [3.8.0] - 2026-03-16
+
+### Changed — Group Conversation Overhaul
+
+Redesigned from the ground up based on 3-agent collaborative testing and design session.
+
+**Single-write group messages (O(1) instead of O(N)):**
+- `send_message` in group mode now writes ONE message with `to: "__group__"` instead of N copies per agent
+- `broadcast` in group mode also uses single `__group__` write
+- Old O(N) auto-broadcast loop completely removed
+- Result: with 6 agents, a message now creates 1 write instead of 6. A broadcast round that previously created 30 writes now creates 6.
+
+**`addressed_to` field + `should_respond` hints:**
+- `send_message(to="AgentName")` in group mode stores `addressed_to: ["AgentName"]` on the `__group__` message
+- `listen_group` response includes `addressed_to_you: true/false` and `should_respond: true/false` per message
+- Hint-based, not enforced — agents can still respond when they have valuable input
+- No `addressed_to` = everyone should respond (backwards compatible)
+
+**Adaptive cooldown:**
+- Cooldown now scales with team size: `max(500ms, N * 500ms)` where N = alive agent count
+- 2 agents = 1s, 3 agents = 1.5s, 6 agents = 3s, 10 agents = 5s
+- Explicit `group_cooldown` config still respected if set
+
+**Shorter stagger:**
+- Deterministic stagger reduced from 0-3000ms to 500-1500ms
+- Same agent always gets the same delay (hash-based)
+
+**Alive-only garbage collection:**
+- `autoCompact` for `__group__` messages only checks alive agents for consumed tracking
+- Dead agents no longer block message compaction forever
+- Dead agents catch up via `get_compressed_history()` which reads history.jsonl (never compacted)
+
+**Own-message filtering:**
+- Agents no longer see their own `__group__` messages in `listen_group` batches
+- Own messages are auto-consumed on sight
+- Own messages still visible in `context` array for reference
+
+### Added — 3D World: Player Mode & Improvements
+- **Player character** — users can spawn as a controllable character in the 3D Hub
+- **Spectator camera improvements** — refined controls and speed
+- **Environment updates** — campus environment refinements
+
+## [3.7.0] - 2026-03-16
+
+### Added — Agent Ecosystem (20 new tools, 52 total)
+
+**Tier 1 — Critical Infrastructure:**
+- **`get_briefing()`** — full project onboarding in one call: agents, tasks, decisions, KB, locked files, progress, project file tree
+- **`lock_file(path)` / `unlock_file(path?)`** — exclusive file editing with auto-release on agent death
+- **`log_decision(decision, reasoning?, topic?)` / `get_decisions(topic?)`** — persistent decision log, prevents re-debating
+- **Agent recovery on rejoin** — `register()` returns active tasks, workspace keys, recent messages for returning agents
+
+**Tier 2 — Quality of Life:**
+- **`kb_write(key, content)` / `kb_read(key?)` / `kb_list()`** — shared team knowledge base (any agent reads/writes)
+- **Event hooks** — auto-fires system messages on `agent_join`, `task_complete`, `all_tasks_done`, `dependency_met`
+- **`update_progress(feature, percent, notes)` / `get_progress()`** — feature-level progress tracking with overall %
+- **`get_compressed_history()`** — auto-compresses old messages into summary segments, keeps recent verbatim
+- **`listen_group()` now blocks indefinitely** — no more timeout, agents never drop out
+
+**Tier 3 — Advanced Collaboration:**
+- **`call_vote(question, options)` / `cast_vote(vote_id, choice)` / `vote_status(vote_id?)`** — team voting with auto-resolve when all vote
+- **`request_review(file, desc)` / `submit_review(review_id, status, feedback)`** — code review pipeline with approve/changes_requested
+- **`declare_dependency(task_id, depends_on)` / `check_dependencies(task_id?)`** — task dependency tracking with auto-notify on resolve
+- **`get_reputation(agent?)` / `suggest_task()`** — agent reputation tracking (auto-detects strengths), task suggestions based on skills
+- **Auto-reputation tracking** — global hook tracks every action (messages, tasks, reviews, decisions, KB writes) without manual calls
+
+### Fixed
+- **Monitor screens stay red** when agent stops listening — persistent color state instead of 300ms flash
+- **"NOT LISTENING" warning** shown prominently on desk monitor canvas
+- **Status color logic** — green = listening, red = active but not listening, yellow = sleeping, dim = dead
+
+## [3.6.2] - 2026-03-16
+
+### Added — Message Awareness System
+- **Sender gets busy status** — `send_message` and `broadcast` tell you when recipients are working (not listening) so you know messages are queued
+- **Pending message nudge** — every non-listen tool call checks for unread messages and tells the agent to call `listen_group()` soon
+- **Message age tracking** — `listen_group` shows `age_seconds` per message and `delayed: true` flag for messages older than 30s
+- **Agent status in batch** — `listen_group` returns `agents_status` map showing who is `listening` vs `working`
+- **listen_group retry** — timeout now returns `retry: true` with explicit instruction to call again immediately
+- **next_action field** — successful `listen_group` response tells agent to call `listen_group()` again after responding
+- **Ctrl key removed from camera** — no longer moves camera down (Q/E only)
+
+### Added — 3D World: Campus Environment & Navigation
+- **Campus environment** — new outdoor environment option with buildings, paths, green spaces
+- **Navigation system** — pathfinding for agents to walk around obstacles instead of through walls
+- **Door animations** — manager office door slides open when agents approach, closes when they leave
+- **Roof visibility** — roof hides when camera is above ceiling height
+
+## [3.6.1] - 2026-03-16
+
+### Fixed
+- **3D Hub black screen on page load** — the office module loads asynchronously, but the initial `switchView('office')` fired before `office3dStart` was defined. Added auto-start at end of module so the 3D Hub loads immediately on refresh.
+
+## [3.6.0] - 2026-03-16
+
+### Added — Managed Conversation Mode
+
+- **`set_conversation_mode("managed")`** — structured turn-taking for 3+ agent teams, prevents broadcast storms
+- **`claim_manager()`** — claim the manager role (first caller wins, auto-election fallback)
+- **`yield_floor(to, prompt?)`** — manager-only: give an agent permission to speak (directed, round-robin `__open__`, or close `__close__`)
+- **`set_phase(phase)`** — manager-only: move team through discussion → planning → execution → review with auto-instructions to all agents
+- **Floor enforcement** — `send_message`, `broadcast`, `handoff`, and `share_file` all block non-floor-holders with actionable error messages
+- **Auto-advance turns** — floor returns to manager after directed responses; round-robin advances to next alive agent automatically
+- **Manager disconnect recovery** — heartbeat detects dead manager within 10-30s, notifies all agents to re-elect
+- **Dead turn-holder detection** — heartbeat detects dead agents holding the floor and resets it
+- **Managed mode in `listen_group()`** — returns `managed_context`, `should_respond`, and `instructions` to guide agent behavior
+- **`managed` template** — 4-agent team (Manager, Designer, Coder, Tester) with structured prompts
+- **`managed-team` conversation template** — dashboard-launchable version
+- **Dashboard Docs tab** — in-dashboard documentation with full tool reference, managed mode guide, architecture, version history
+- **Dashboard managed mode badge** — header shows current phase and floor status when managed mode is active
+
+### Added — 3D World Improvements
+
+- **Spectator camera** — free-fly WASD + mouse camera replacing OrbitControls, no distance limits, Shift for fast movement, Q/E up/down
+- **6 new hairstyles** — curly, afro, bun, braids, mohawk, wavy
+- **6 new eye styles** — surprised, angry, happy, wink, confident, tired
+- **5 new mouth styles** — grin, frown, smirk, tongue, whistle
+- **6 outfit types** — hoodie, suit, dress, lab coat, vest, jacket with color customization
+- **3 body types** — default, stocky, slim (scale multipliers on torso/legs/arms)
+- **5 gesture animations** — wave, think, point, celebrate, stretch with idle gesture system
+- **New furniture** — bookshelf (random colored books), wall TV (animated dashboard with agent stats, scrolling ticker, clock), arcade machine (cabinet + screen + joystick + buttons), floor lamp (warm point light), area rug
+- **Agent behavior** — realistic conversation distance (1.8m), listener turns toward speaker, broadcast triggers wave gesture, task completion triggers celebrate
+- **3D Hub** — renamed from "Office", now default tab on page load
+- **Speed slider** — camera speed control in toolbar (1-20)
+
+### Added — 3D Virtual Office (v1 foundation from previous session)
+
+- **Modular 3D engine** — 14 ES modules under `office/`
+- **Expanded office** — 28x16 floor with right wing, dividing wall, LOUNGE archway
+- **Dressing room** — mirror, raised platform, privacy partitions, coat hooks
+- **Rest area** — beanbags, circular rug, side table, warm ambient lighting
+- **Click-to-command** — Dressing Room, Go Rest, Back to Work, Edit Profile
+- **Character designer** — 5-tab panel with live 3D rotating preview
+- **Accessory system** — glasses, headwear, neckwear with color customization
+- **Mod system infrastructure** — GLB/GLTF pipeline with validation
+
+### Security
+- **Config file lock** — `config.json` read-modify-write operations now use file-based locking (same pattern as `agents.json`)
+- **Reserved name blocklist** — `__system__`, `__all__`, `__open__`, `__close__`, `system` cannot be registered as agent names
+- **Mode change protection** — only the manager can switch away from managed mode
+- **Floor enforcement on all message paths** — `handoff` and `share_file` now enforce managed mode floor control
+- **Branch-aware system messages** — floor/phase notifications sent to recipient's branch, not sender's
+- **Phase history cap** — limited to 50 entries to prevent config.json bloat
+- `/office/*` and `/mods/*` static routes with path traversal protection
+- Mod file type allowlist blocks all executable formats
+- GLB magic bytes validation (server + client)
+
+### Removed
+- ~1,100 lines of dead 2D isometric office code
+
+## [3.5.0] - 2026-03-15
+
+### Added — Group Conversation Mode
+- **`set_conversation_mode("group")`** — enables free multi-agent collaboration with auto-broadcast
+- **`listen_group()`** — batch message receiver with random stagger (1-3s) to prevent simultaneous responses
+- Returns ALL unconsumed messages + last 20 messages of context + hints about silent agents
+- Auto-broadcast in group mode: every message is shared with all agents automatically
+- Cooldown enforcement: agents must wait 3s between sends to maintain conversation flow
+- Cascade prevention: broadcast copies don't trigger further broadcasts
+- MCP tools: 27 → 29
+
+### Added — Dashboard Features
+- **Notification panel** — bell icon with badge count, dropdown event feed (agent online/offline, listening status changes)
+- **Agent leaderboard** — performance scoring (0-100) with responsiveness, activity, reliability, collaboration dimensions
+- **Cross-project search** — "All Projects" toggle in search bar, searches across all registered projects
+- **Animated replay export** — Export conversation as self-playing HTML file with typing animations and play/pause controls
+- **Ollama integration** — `npx let-them-talk init --ollama` auto-detects Ollama, creates bridge script for local models
+
+### Fixed — PID & Registration Integrity
+- Registration file locking with try/finally (prevents race conditions when multiple agents register simultaneously)
+- PID stale detection uses `last_activity` with 30s threshold (prevents false "alive" from Windows PID reuse)
+- Lock file cleaned up on process exit
+- Dashboard inject/nudge snapshots project context at click time (prevents wrong-project race)
+
+### Security
+- `toolHandoff` and workflow auto-handoff now check `canSendTo` permissions
+- `lastSentAt` updated in `toolBroadcast` (prevents cooldown bypass)
+- `config.json` added to both server and dashboard reset cleanup
+- Auto-broadcast respects `canSendTo` per recipient
+
+## [3.4.4] - 2026-03-15
+
+### Fixed
+- Add project now accepts any existing directory (removed requirement for package.json or .git)
+- Init safely backs up corrupted .mcp.json and settings.json before overwriting
+
+### Changed
+- Removed plugin references from website and docs
+- Website updated with security features (LAN auth token, CSRF, CSP)
+
+## [3.4.3] - 2026-03-15
+
+### Removed — Plugin System
+- Removed the entire plugin system (`vm.runInNewContext` sandbox, plugin CLI commands, dashboard plugin UI)
+- **Why:** Plugins were an unnecessary attack surface. Node.js `vm` is not a security sandbox — plugins could escape and execute arbitrary OS commands. CLI terminals (Claude Code, Gemini, Codex) have their own extension systems, making our plugins redundant.
+- `npx let-them-talk plugin` now shows a deprecation notice
+- MCP tools reduced from 27 + plugins to 27 (all core tools remain)
+- ~200 lines of code removed from server.js, cli.js, dashboard.js, dashboard.html
+
+## [3.4.2] - 2026-03-15
+
+### Security — CSRF Protection
+- Required `X-LTT-Request` custom header on all POST/PUT/DELETE requests
+- `lttFetch` wrapper in dashboard automatically includes the header
+- Malicious cross-origin pages cannot set custom headers without CORS preflight approval
+- Removed wildcard `Access-Control-Allow-Origin: *` in LAN mode — now uses explicit trusted origins only
+- Empty Origin/Referer no longer auto-trusted — requires custom header as minimum protection
+
+### Security — LAN Auth Token
+- Auto-generated 32-char hex token when LAN mode is enabled
+- Token required for all non-localhost requests (via `?token=` query param or `X-LTT-Token` header)
+- Token included in QR code URL — phone scans and it just works
+- Token displayed in phone access modal with explanation
+- New token generated each time LAN mode is toggled on
+- Token persists across server restarts via `.lan-token` file
+- Localhost access never requires a token
+
+### Security — Content Security Policy
+- CSP header added to dashboard HTML response
+- `script-src 'unsafe-inline'` for inline handlers, blocks `eval()` and external scripts
+- `connect-src 'self'` restricts API calls to same origin
+- `font-src`, `style-src`, `img-src` scoped to required sources only
+
+### Fixed
+- CSRF brace imbalance that trapped GET handlers inside POST-only block
+- LAN token not forwarded from phone URL to API calls and SSE
+- Redundant nested origin check collapsed to single condition
+
+## [3.4.1] - 2026-03-15
+
+### Added
+- **File-level mutex** — in-memory promise queue per file for serializing edit/delete operations
+- **Agent permissions enforcement** — `canSendTo()` checks in `send_message` and `broadcast`, `can_read` filtering in `get_history` and message delivery
+- **Read receipts** — auto-recorded when agents consume messages, visible as agent-initial dots under messages in dashboard
+
+### Security
+- HTTP 500 responses now return generic error instead of raw `err.message` (prevents filesystem path leaks)
+- `/api/discover` changed from GET to POST (now under CSRF protection)
+- `workspace_read`/`workspace_list` validate agent name parameter with regex
+- `get_history` filters results by agent's `can_read` permissions
+- `read_receipts.json` and `permissions.json` added to both MCP and dashboard reset cleanup
+- Dashboard workspace API regex aligned with server (`[a-zA-Z0-9_-]`)
+
+### Fixed
+- `toolWaitForReply` missing `markAsRead` calls (read receipts not recorded)
+- `toolBroadcast` bypassing permission checks entirely
+- `toolReset` not cleaning up `permissions.json` and `read_receipts.json`
+
+## [3.4.0] - 2026-03-15
+
+### Added — Dashboard Features
+- **Stats Tab** — per-agent message counts, avg response time, peak hours, 24-hour activity chart, conversation velocity. Keyboard shortcut `6`.
+- **Compact View** — toggle button in search bar. Hides avatars, inlines timestamps, reduces padding. Persists to localStorage.
+- **Message Edit** — edit any message via hover action. Full edit history tracked, "edited" badge displayed.
+- **Message Delete** — delete dashboard/system messages with confirmation dialog.
+- **Copy Message** — clipboard button on message hover to copy raw content.
+- **JSON Export** — new export format alongside HTML and Markdown.
+- **Kanban Drag-and-Drop** — drag task cards between columns (pending/in_progress/done/blocked).
+- **SSE Auto-Reconnect** — exponential backoff (1s→30s), yellow "Reconnecting..." indicator, polling fallback.
+- **Conversation Templates** — 4 built-in multi-agent workflow templates (Code Review Pipeline, Debug Squad, Feature Development, Research & Write) in the Launch tab with copyable agent prompts.
+
+### Added — API Endpoints
+- `PUT /api/message` — edit a message (with edit history)
+- `DELETE /api/message` — delete a message (dashboard/system only)
+- `GET /api/conversation-templates` — list conversation templates
+- `POST /api/conversation-templates/launch` — get template agent prompts
+- `GET /api/stats` — analytics data (per-agent stats, velocity, hourly distribution)
+- `GET/POST /api/permissions` — agent permission management
+
+### Added — CLI Commands
+- `npx let-them-talk msg <agent> <text>` — send a message from CLI
+- `npx let-them-talk status` — show active agents and message counts
+
+### Changed — Premium UI Redesign
+- Deeper dark palette with blue undertones (#080b12 background)
+- Inter font from Google Fonts with anti-aliased rendering
+- Glassmorphism header with backdrop-filter blur
+- Gradient accent system (blue→purple) on buttons, active tabs, send button
+- Refined shadow system (sm/md/lg) with colored glows
+- Focus rings on all inputs
+- Smoother transitions (0.2-0.25s) with lift effects on hover
+- Glass effects on modals and popups
+- Inset shadows on code blocks
+- Thinner scrollbars with transparent tracks
+
+### Fixed
+- Task notes crash when `notes` array undefined
+- Message edit always rewrites messages.jsonl regardless of match
+- Permissions API accepted arbitrary fields (now whitelisted)
+- Task status accepted any string (now validated against whitelist)
+- Reset button ignored active project in multi-project mode
+- Edit modal missing error handler on network failure
+- CLI msg command accepted invalid agent names
+- Copy-to-clipboard double-escaped HTML entities in template prompts
+- Duplicate deleteMessage function shadowing
+
+## [3.3.2] - 2026-03-14
+
+### Changed
+- License changed from MIT to Business Source License 1.1 (BSL)
+- Added SECURITY.md with vulnerability disclosure policy
+- Added CHANGELOG.md to published npm package
+- Added .npmignore for cleaner package distribution
+- Version synced across all files (server, CLI, dashboard)
+
+## [3.0.0] - 2026-03-14
+
+### Added — Agent Profiles
+- New tool: `update_profile` (display_name, avatar, bio, role)
+- 12 built-in SVG robot avatar icons with hash-based defaults
+- Profiles auto-created on register, persist across restarts
+- Profile data shown in dashboard (avatars, role badges, profile popup)
+
+### Added — Agent Workspaces
+- 3 new tools: `workspace_write`, `workspace_read`, `workspace_list`
+- Per-agent key-value storage (50 keys max, 100KB per value)
+- Agents can read anyone's workspace, write only their own
+- Dashboard "Workspaces" tab with collapsible accordion UI
+
+### Added — Workflow Automation
+- 3 new tools: `create_workflow`, `advance_workflow`, `workflow_status`
+- Multi-step pipelines with auto-handoff to step assignees
+- Dashboard "Workflows" tab with horizontal pipeline visualization
+- Dashboard can advance/skip workflow steps
+
+### Added — Conversation Branching
+- 3 new tools: `fork_conversation`, `switch_branch`, `list_branches`
+- Fork at any message point with isolated branch history
+- All message tools branch-aware (backward compatible — main branch uses existing files)
+- Branch tabs in dashboard
+
+### Added — Plugin System
+- Dynamic tool loading from `plugins/*.js` files
+- Sandboxed execution with 30s timeout
+- CLI: `npx let-them-talk plugin add/list/remove/enable/disable`
+- Dashboard plugin cards with enable/disable toggles
+
+### Changed
+- MCP tools: 17 → 27 + dynamic plugins
+- Dashboard tabs: 2 → 4 (Messages, Tasks, Workspaces, Workflows)
+- Branch-aware history API (`?branch=` query param)
+- Version bump across all files (server, dashboard, CLI, package.json)
+
+## [2.5.0] - 2026-03-14
+
+### Added
+- Task management system: `create_task`, `update_task`, `list_tasks` tools
+- Kanban board in dashboard (Messages/Tasks toggle)
+- Agent stats panel (sent/received/avg response time per agent)
+- Shareable HTML export (/api/export endpoint)
+- Export dropdown (HTML + Markdown formats)
+- Conversation bookmarks (star messages, localStorage)
+- Sound notification toggle (Web Audio API)
+- Typing indicator for processing agents
+- Connection quality display (SSE latency)
+- Date separators between message groups
+- Message grouping for consecutive same-sender messages
+- Project auto-discover (scan nearby folders)
+- Copy-to-clipboard prompts in onboarding
+- Dynamic tab title with message count
+- Dashboard footer with version
+
+### Security
+- Path traversal fix in `share_file` (restricted to project dir)
+- Path traversal fix in `?project=` param (validate against registered projects)
+- 1MB message size limit on send/broadcast/handoff
+- 1MB request body limit on dashboard POST endpoints
+- XSS fix in HTML export (escape agent names)
+- CORS restricted to localhost only (was wildcard)
+- Dashboard binds to 127.0.0.1 only (was 0.0.0.0)
+- Registration guard on `reset` tool
+- Removed absolute file paths from share_file responses
+
+## [2.3.0] - 2026-03-14
+
+### Added
+- `handoff` tool for structured work delegation
+- `share_file` tool for sending file contents between agents
+- `broadcast` tool for messaging all agents at once
+- `get_summary` tool for conversation recaps
+- Server-Sent Events for real-time dashboard updates
+- `fs.watch()` on data directory with debounced SSE push
+- Graceful SSE fallback to polling
+- Handoff message rendering (purple banner)
+- File share message rendering (file icon + size)
+
+## [2.1.0] - 2026-03-14
+
+### Added
+- Multi-agent support (any name, not just A/B)
+- `list_agents` tool with alive/dead status
+- `listen` tool (blocks indefinitely, never times out)
+- Conversation threading (`reply_to` + auto `thread_id`)
+- Message acknowledgments (`ack_message` tool)
+- Heartbeat system (10s interval, `last_activity` tracking)
+- Agent status: active/sleeping/dead with idle time
+- Listening status tracking (`listening_since`)
+- Auto-compact messages.jsonl when >500 lines
+- Auto-archive conversations before reset
+- Context hints when conversation exceeds 50 messages
+- Dead recipient warnings in `send_message`
+- Message sequence numbers for ordering
+- `pending_count` and `agents_online` in delivery responses
+- 4 agent templates: pair, team, review, debate
+- CLI: `npx let-them-talk templates` command
+- CLI: `--template` flag for guided setup
+- Multi-CLI support: Claude Code, Gemini CLI, Codex CLI
+- `AGENT_BRIDGE_DATA_DIR` env var in MCP config
+
+### Fixed
+- Heartbeat timer `.unref()` to prevent zombie processes
+- Process exit cleanup (deregister agent on exit)
+- Re-registration cleanup (old name removed)
+- Stale byte offset recovery on file truncation
+
+## [2.0.0] - 2026-03-14
+
+### Added
+- Initial release
+- MCP server with stdio transport
+- 6 tools: register, send_message, wait_for_reply, check_messages, get_history, reset
+- Web dashboard with real-time monitoring
+- Message injection from dashboard
+- Dark theme UI with markdown rendering
+- `.mcp.json` project-level configuration