Paperclip Updates Default Agent Runner, Adds Native Session Recovery and Migration Checks

The open-source agent management tool Paperclip released v2026.1106.0 on November 6, making the native Paperclip Runner the default execution path for some newly created agents, including Claude, Codex, OpenCode, and Pi. Existing agents do not switch automatically after an upgrade. Eligible Claude and Codex agents can be converted with “Upgrade runner,” which also shows settings that cannot be carried over. Release announcement

The key technical change is to integrate the execution lifecycle, built-in task tools, and native session identification into the Runner, then select the model provider and execution environment separately. When resuming a session, the system rechecks the provider, credentials, environment, and task permissions. However, recovery capabilities vary by provider, and saving a session does not guarantee that every interrupted operation can be resumed. For operations teams, this makes the checks for session recovery clearer, while making provider differences something that must be verified during deployment. Runner documentation

The default switch has specific platform requirements: supported systems include Intel and Apple Silicon Macs, as well as Linux x64/ARM64 with GNU libc 2.39 or later. OpenCode and Pi still use the old runner on Linux ARM64. Windows, Alpine and other musl-based systems, and environments with older libc versions also retain the existing execution path. Eligibility is determined by the host where the agent actually runs, so container images and remote hosts must be checked separately. Platform requirements

Migration may also encounter incompatible settings. Some custom commands, extra parameters, file or network scopes, and turn limits from older adapters have no equivalent in the Runner. The system will reject the save and list the conflicting fields. Engineering teams should test the environment and permissions first, then verify interrupted-session recovery, usage statistics, and the behavior of existing scripts. A successful installation alone is not enough to establish production readiness. Migration limitations

This release also adds native execution support for GitHub Copilot and changes the operations interface: the execution history API now returns 200 records by default, with a maximum of 1,000, so clients that rely on unbounded arrays must update their queries. Agent execution failures are also no longer sent to Sentry by default. These changes can affect retrieval of historical data and alert coverage after an upgrade, and should be included in deployment acceptance checks. Breaking changes