Skip to content

v1.126.0

Latest
Latest

Choose a tag to compare

View all tags
@docker-read-write docker-read-write released this 18 Aug 20:46
· 2 commits to main since this release
v1.126.0
1719b31
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

This release adds evaluation regression gating, session comparison for replay, and significant hardening of served-agent safety controls across A2A, MCP HTTP, and chat surfaces.

What's New

  • Adds --baseline and --regression-tolerance flags to docker agent eval, enabling CI-style regression detection by comparing a run against a previously saved baseline and failing on quality regression
  • Adds docker agent sessions diff command to compare the behavior of two recorded sessions, reporting the first point of divergence between runs
  • Adds unattended tool execution restrictions for A2A, requiring explicit security configuration for network listeners
  • Adds centralized unattended safety resolution for served agents, with ordered concrete safety policies and hardened HTTP serving for MCP and chat surfaces
  • Adds session-origin isolation for A2A session resumes

Bug Fixes

  • Fixes the baseline gate to be grounded on the file that eval writes, and fails closed on error
  • Fixes replay comparison to handle sub-agent turns, compare arguments semantically, and sanitize output
  • Fixes A2A session resumes to be isolated by origin

Technical Changes

  • Refactors sessions diff CLI command to resolve session references
  • Refactors safety flag validation to be shared across CLI surfaces
  • Refactors HTTP auth and origin matchers into a shared httpsec package
  • Refactors A2A server assembly, extracting it from Run
  • Replaces CAGENT_MODELS_GATEWAY with DOCKER_AGENT_MODELS_GATEWAY in docs and samples
  • Documents the evaluation regression gate and its two flags, the docker agent sessions diff command, and served-agent safety controls

What's Changed

  • docs: update CHANGELOG.md for v1.125.0 by @docker-read-write[bot] in #3995
  • feat(replay): compare the behaviour of two recorded sessions by @dwin-gharibi in #3948
  • feat(eval): compare a run against a saved baseline and fail on regression by @dwin-gharibi in #3946
  • docs: replace CAGENT_MODELS_GATEWAY with DOCKER_AGENT_MODELS_GATEWAY in docs and samples by @aheritier in #3997
  • feat: harden served-agent safety controls by @aheritier in #4000

Full Changelog: v1.125.0...v1.126.0

Contributors

aheritier and dwin-gharibi
Assets 8
Loading