Skip to content

CtrlNode Bridge 2026.2.5

Latest
Latest

Choose a tag to compare

View all tags
@ctrlnode-ai ctrlnode-ai released this 26 Jun 23:57
v2026.2.5
a439bca
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

CtrlNode Bridge v2026.2.5 Release Notes

This release adds task cancellation, user input delivery, and persistent session resume for follow-ups across all providers.

What's new

Task cancellation

Tasks can now be cancelled mid-run from the CtrlNode UI. Bridge receives a cancel_task command and signals the active provider to stop immediately:

  • Claude Code / Hermes CLI — sends SIGTERM to the child process, then SIGKILL after 3 seconds if still running
  • Claude Agent SDK — aborts via AbortController
  • Codex SDK — sets a cancel flag that breaks the event stream loop cleanly
  • Copilot / Gemini / Hermes ACP — sends SIGTERM to the ACP subprocess

User input delivery

When an ACP agent (Copilot, Gemini, Hermes ACP) encounters a permission prompt it cannot auto-approve, it now notifies the UI with a task_waiting_for_input event instead of silently cancelling. The UI can respond with an input_response message that Bridge forwards to the running process. Claude Code supports writing directly to stdin; ACP providers log a warning for now (full interactive support coming in a future release).

Persistent session resume for follow-ups (Claude Agent SDK)

Follow-up intents dispatched to Claude Agent SDK agents now resume the original conversation session instead of starting a new one:

  • The session ID is saved to disk (tasks/<taskId>/session_id) when a task completes so it survives Bridge restarts
  • On follow-up, Bridge reads the session ID from memory or disk and passes it to the SDK for full conversation history resumption
  • A fresh dispatch (e.g. RERUN) always clears the cached session so it never accidentally resumes a stale one

Follow-up file preparation (all providers)

Follow-up messages now consistently write a numbered input file (<shortId>-followup-N.md) to the task's input/ folder before dispatching to any provider. The agent prompt is automatically augmented with:

  • An instruction pointing to the follow-up output file the agent should write (<shortId>-followup-N-output.md)
  • For stateless providers (Codex, Hermes CLI/ACP, Gemini, Copilot): the prior agent_log.md is prepended as context so the agent knows what was done in the original run

Providers with native session history (Claude Code, Claude Agent SDK) receive only the output-file instruction since the conversation history is already available in the session.

Follow-up emits task_complete

Follow-up runs now emit a task_complete event when they finish, causing the backend to properly transition the task status back to done or failed after the follow-up completes.

Upgrade

Replace the binary and restart. No configuration changes are required.

Assets 8
Loading
0 Join discussion