Starstruk Logo MarkStarstruk

v0.115.3

Agenta-AI/agentav0.115.3Sep 9, 2026by mmabrouk

AI Summary

This release focuses on enhancing agent capabilities by introducing platform instructions and custom secrets, alongside critical fixes for session handling and frontend rendering loops. It also includes updates to agent permissions, tool sets, and documentation for self-hosted setups.

Key Highlights

  • Added platform instructions and custom secrets for agents
  • Simplified agent permissions and shipped the agent tool set
  • Fixed session watchdog failures and context preservation
  • Resolved frontend rendering loops in the 'Use API' drawer
  • Updated self-hosting documentation for API runner URLs

New Features

  • Add platform instructions and custom secrets
  • Simplify agent permissions and allow routine work by default
  • Ship the agent tool set in runner images and Daytona snapshot
  • Give every turn current agent and session context

Full Release Notes

## What's Changed
* fix(sessions): preserve active turns when watchdog checks fail by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6615
* [fix] Update attachment acceptance test for saved chat records by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6621
* [AGE-4067] feat(agents): Add platform instructions and custom secrets by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6365
* [6629] docs(self-host): the api needs the runner URL for Stop, plus the matching troubleshooting entry by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6630
* [6627] fix(hosting): give the api service the runner URL on the gh.local and gh.ssl stacks by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6628
* feat(agents): rewrite the platform prompt and ship request_secret in the build kit by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6632
* feat(agents): align the build-an-agent skill with the platform prompt by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6633
* feat(agents): give every turn current agent and session context by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6638
* [feat] Ship the agent tool set in both runner images and the Daytona snapshot by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/5796
* [feat] Simplify agent permissions and allow routine work by default by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6641
* fix(frontend): hold the idle deferral in the desktop permissions test by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6654
* fix(api): make ee unit tests importable as a package by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6656
* [fix] Open the file link the platform prompt tells the agent to write by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6665
* [test] Drop the partial escape CodeQL flagged in the link-gate test by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6670
* [6669] test(gate): Rename a session between turns and assert the agent was told by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6673
* [AGE-4290] fix(sdk): Read the session facts in the agent service, not off the wire by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6667
* [fix] Stop the Permissions dropdown claiming the wrong default by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6672
* [fix] Refuse a chat link whose target names a host, not a path by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6675
* [chore] Let the session-context gate cell choose its sandbox by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6678
* [docs] Say which permission policy a new agent really starts on by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6681
* [fix] Stop telling agents that allow_reads is the new-agent default by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6682
* [chore] Correct three test comments that claimed more than the tests prove by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6687
* [test] Wait for the Use API drawer to settle before clicking inside it by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6685
* [fix] Read provider-key presence from value_status, not the write-only value by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6691
* [6683] fix(frontend): make the Rename menu entry open the inline editor by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6686
* fix(frontend): show a sent chat message at once instead of after the record read by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6658
* [fix] Refuse an automatic session rename over a name a person typed [AGE-4287] by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6676
* [AGE-4302] fix(frontend): Stop offering a custom provider under a harness that cannot reach it by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6694
* fix(frontend): drop a refused message's row only once the composer confirms it took the text by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6705
* [test] Quarantine the variant Use API test on the drawer freeze [#6708] by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6709
* [fix] Stop the Use API drawer render loop with two variants [#6708] by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6711
* [test] Re-enable the Use API acceptance test after the drawer loop fix by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6713
* [test] Replace every newline in the CRLF framing test (CodeQL) by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/6719
* [release] v0.115.3 by @github-actions[bot] in https://github.com/Agenta-AI/agenta/pull/6616


**Full Changelog**: https://github.com/Agenta-AI/agenta/compare/v0.115.2...v0.115.3