Self-recovery: dispatch dict, disable updater in Docker, fix watchdog semantics ...

1. Replace if/elif dispatch chain with _DISPATCH dict (engines/node.py)
   Adding a proto message kind now requires exactly one entry in _DISPATCH —
   impossible to add a callback slot without the routing case. Missing kinds
   emit WARNING immediately. Extractor is colocated with the route so the
   wrong-object bug (Bug 1) cannot recur.

2. Honour CORTEX_DISABLE_AUTO_UPDATE env var (updater.py)
   In Docker, updates are delivered via image rebuilds. The old behaviour
   called sys.exit(0) to spawn a bootstrapper, which Docker restart:always
   turned into an infinite boot loop. Setting CORTEX_DISABLE_AUTO_UPDATE=1
   in the container env prevents this entirely.

3. Watchdog ticks unconditionally in health reporter (node.py)
   Previously the watchdog tick was skipped when the hub was unreachable,
   causing os._exit(1) after 300s of any disconnect — even during normal
   gRPC reconnect backoff. Now the watchdog proves the reporter thread is
   alive regardless of hub state; it only fires if the thread itself deadlocks.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Fix node disconnect: health stream restart, policy dispatch, work_pool routing ...

Three bugs causing nodes to drop offline and stay offline after disconnect:

1. on_policy passed full ServerMessage instead of sub-message — caused
   Dispatch Error: mode on every connection (AgentNode._on_policy_update
   accesses policy.mode directly)

2. _health_thread_started never reset in close() — health stream could not
   restart after reconnect, so hub watchdog eventually timed out the node

3. work_pool_update had no dispatch case in on_message() — _handle_work_pool
   was dead code, causing hub to flood nodes with unclaimed task updates

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Fixed node crash due to missing _apply_skill_config and added debug logs

Fix AgentNode crash on Windows and fix bootstrap extraction nesting bug

Add token self-recovery to survive auth failures without SSH access ...

- GrpcMeshTransport: on handshake rejection, calls /api/v1/agent/token-sync
  using stable secret_key to fetch fresh invite_token and retries handshake.
  Persists recovered token to all known config file locations.
- agent_update.py: new /token-sync endpoint (auth: hub SECRET_KEY header).
- node.py: wire hub_http_url + secret_key into GrpcMeshTransport constructor.
- reinstall_windows_agent.ps1: idempotent all-in-one reinstall script —
  kills competing python processes, disables ghost startup bat, syncs config,
  optionally updates token, re-registers task with RestartCount=3.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Fixes for integration tests in mocked mode and node deadlock

Fix race condition in auto-healing transport loop

Implement fundamental auto-healing for disconnected transports

Fix AttributeError on TaskCancelRequest and include symlinks in update package

Fix integration tests, deadlocks, and race conditions in coworker flow

refactor done

half done refactoring

cleanup

fix: stabilize integration tests, restore admin bypass, and implement memory safety caps

Hardened Windows Agent Task Termination and Automatic Cleanup

feat: complete Windows swarm integration and remote terminal functionality

feat(mesh): autonomous windows provisioning and lifecycle management ...

- Implemented secure self-purge mechanism for agent nodes (purge.py).
- Refactored bootstrap_windows.ps1 for headless/non-interactive installation.
- Added POST /nodes/purge endpoint to Hub for token-based self-deregistration.
- Enhanced Nodes UI with 'Purge & Deregister' option and platform-aware command dispatch.
- Modularized backend services (Mesh, NodeRegistry, Session, Tool, RAG) to centralize business logic.
- Stabilized Windows PTY serialization and terminal output dispatch.
- Added integration tests for node purge flows.

Fix Windows PyWinPTY terminal boundary string drops via bat spool execution and full Windows deployment support

chore: Stable Agent Mesh - Deterministic HMAC Signatures and Proto Synchronization ...

- Fix circular import in Agent auth utility (Live config vs static import)
- Enforce deterministic Protobuf serialization for cross-platform signature consistency
- Synchronize stale Protobuf classes in agent-node source tree
- Sign gRPC keepalive/work-pool messages to prevent stream drops
- Fix Hub config initialization order (OIDC_ENABLED vs ALLOW_PASSWORD_LOGIN)
- Enhanced stream error logging with tracebacks
- Bumped version to 1.1.0 to force fleetwide update

Remediation of audit reports (gRPC Security, Node Auth, and Protos)

Auto-cleanup dead systemlink & FS Cache Reflowing, Bump OTA Version to 1.0.76

fix(sync): memory exhaust on large file mesh broadcast by enforcing synchronous stream backpressure, increasing chunk size to 4MB, and bounding priority queues

perf: implement local file metadata hash cache for manifest generation and reconciliation

fix: resolve zero byte file sync breaking push loop iteration

fix: resolve file sync infinite bounce loop and unbound local variable bugs

fix: cat() on deleted file returns instantly instead of timing out ...

Root cause: two separate silent-failure bugs caused 15s stalls on file-not-found:

1. agent-node/node.py _push_file(): when a READ arrives for a missing file,
   silently returned without sending any response. The Hub journal sat waiting
   for the full 15s timeout.
   Fix: send ERROR SyncStatus back immediately with task_id so the journal
   resolves at once.

2. ai-hub/assistant.py cat(): after the mirror fast-path found no file, it
   fell through and sent a READ gRPC to the node anyway, compounding the wait.
   Fix: short-circuit with {error: File not found} when the file is absent
   from the session mirror — no node round-trip needed.

test: added Case 9 (TestSmallFileSync) — asserts that cat() on a freshly
deleted file resolves in < 3s on both nodes (guard against regression).

fix(file_sync): check protobuf DESCRIPTOR to correctly detect total_chunks support for fast-path processing

Refactor browser service for parallel worker pool and deep research capabilities. Clean up redundant directories and fix potential path issues for .browser_data.

Fix AI loop stalling and large file sync issues

feat(agent): implement dynamic configuration reloading ...

- The Agent Node now actively reloads its config file during the heartbeat retry loop. \n- Supports both original flat YAML and nested 'cortex' token YAML.\n- Agent no longer caches its node_id or auth_token dynamically at startup, drastically improving resilience on edge.

feat(agent): implement resilient auth handshake loop ...

- The Agent Node will no longer crash and exit if the initial token validation fails or the Orchestrator API is unavailable.\n- Replaced sys.exit(1) on SyncConfiguration rejection with a 5-second sleep and retry loop.\n- This allows nodes to transparently recover once a user dynamically updates their background config with the correct token.

fix(mesh): resolve synology shell bug + bandwidth throttle ...

- Changed mesh-terminal-control to fallback to /bin/sh if bash is missing.\n- Added ~25MB/s bandwidth throttle to Hub/Node file sync to prevent gRPC streaming timeouts.\n- Bumped agent version to 1.0.20.

feat(agent): support FS_ROOT awareness in chat sessions to fix Nas access ...

- Updated Agent to allow out-of-session file access if path is within FS_ROOT.\n- Bumped agent version to 1.0.18.

refactor: organize agent-node client code structure by moving core packages to src/ ...

- Moved agent_node, protos, and shared_core inside src/

- Updated Dockerfile, deployment scripts, compile_protos.sh, installers

- Updated bundle generation in ai-hub to reflect cleaner structure and better README