Skip to content

CHANGELOG.md

Latest commit

 

History

History
43 lines (27 loc) · 2.9 KB

CHANGELOG.md

File metadata and controls

43 lines (27 loc) · 2.9 KB
Raw

Changelog

0.1.0 - 2026-06-15

Breaking Changes

  • Initial Release: 0.1.0 is the first public release of AgentGate. There is no upgrade path from an earlier stable version yet; this version defines the initial CLI, dashboard, AgentPack, and artifact contract.

Major Features and Improvements

  • What AgentGate Is:

    • Release Authority for AI Agents: AgentGate is a pre-production release gate for AI agents. It reads Phoenix trace evidence, applies policy and release metrics, and answers one question: can this candidate version safely ship?
    • Deterministic Decision Path: The first version establishes a reproducible APPROVED / BLOCKED decision flow driven by evidence and policy thresholds rather than human intuition alone.
    • Auditable Output: Every release check produces an audit bundle so teams can inspect why a candidate passed or failed.

  • Core Product Capabilities:

    • Phoenix Evidence Source: AgentGate can run release checks against Phoenix-backed traces, spans, eval labels, policy preflights, and tool activity.
    • Local JSONL Fallback: The first version also supports offline evidence replay for demos, testing, and reproducible review.
    • Release Reports: Added HTML and JSON release artifacts including release decision, metrics summary, dangerous sessions, regression gates, and verification results.

  • AgentPack Model:

    • Two-Layer EffectiveConfig: The product ships with a PhoenixBase plus AgentCustom configuration model for policies and metrics.
    • DefaultDemoPack: configs/agents/stability_ops/ serves as the bundled reference AgentPack for the first public version.
    • Agent-Agnostic Boundary: Production agent behavior stays outside AgentGate. AgentGate integrates through AgentPack config, trace evidence, and tool registration.

  • Operator Experience:

    • Web Dashboard: Added landing, run, latest report, artifact download, and health routes for reviewing release status.
    • Future Controls: Blocked candidates can emit release controls that later candidates must verify.
    • Optional AI Assistance: Advisory review and dangerous-session diagnosis are available without taking release authority away from deterministic policy evaluation.

  • Foundation for Future Versions:

    • Testing Coverage: The first release includes unit coverage for release checks, dashboard rendering, AgentPack loading, Phoenix normalization, telemetry replay, and demo seed generation.
    • Maintainable Architecture: The codebase establishes the initial AgentPack, release, reporting, and dashboard seams for future product work.

Bug Fixes and Other Changes

  • Version Metadata Alignment: Standardized project metadata on version 0.1.0 across Python and Node package manifests.
  • Validation Workflow: Verified the release with unit tests, Ruff checks, AgentPack config validation, profile validation, and suite validation.