Skip to content

CHANGELOG.md

Latest commit

 

History

History
80 lines (67 loc) · 4.19 KB

CHANGELOG.md

File metadata and controls

80 lines (67 loc) · 4.19 KB

Changelog

v0.2.2-alpha

  • add a LangGraph-backed agent runtime adapter that traces real model and tool execution into AgentLens events
  • add a real LangGraph demo wired to OpenAI-compatible providers through LangChain create_agent
  • add shareable trace bundle export so a run can be zipped with raw trace, HTML view, diff, and summary manifest
  • strengthen the OpenAI-compatible wrapper to support direct SDK calls, provider base_url, and better response normalization
  • update README, roadmap, and next-step docs so the published project state matches the actual implementation

v0.2.1-alpha

  • add failure chains and answer-risk labels to make failure explanation more actionable
  • add divergence timelines and severity labels to make run comparison more useful
  • extend viewer summary to show failure mode and chain evidence

v0.2.0-alpha

  • add golden-output tests for diff and regression report snapshots
  • add malformed-input coverage for trace loading, wrapper fallbacks, viewer rendering, and baseline failure paths
  • move the validation strategy beyond self-authored happy-path tests toward output contracts and adversarial cases

v0.1.9-alpha

  • extend regression workflow tests to cover baseline listing and report contents
  • extend viewer tests to assert more of the highlighted debugging controls
  • keep tightening coverage around the developer-facing workflows added in recent iterations

v0.1.8-alpha

  • add deeper regression coverage for baseline persistence and wrapper instrumentation
  • add explicit tests for span error emission and OpenAI-compatible wrapper metrics capture
  • strengthen confidence that the newer integration and regression workflows behave as expected

v0.1.7-alpha

  • tighten the README opening around the local-first debugging wedge
  • add go-to-market notes covering positioning, launch channels, and demo hooks
  • add first-pass launch copy drafts for GitHub, Reddit, X, and Show HN
  • strengthen the external narrative around failure explanation, memory attribution, and regression diffing

v0.1.6-alpha

  • add a minimal OpenAI-compatible wrapper demo for lower-friction LLM tracing
  • add local baseline save/list commands and Markdown regression reports
  • extend the CLI with demo openai-wrapper, baseline, and regression check
  • document integration and regression workflows in the README

v0.1.5-alpha

  • add optional privacy-safe redaction for local traces
  • redact common secret-like strings, emails, phone numbers, and configured sensitive keys before writing JSONL events
  • expose redaction helpers from the SDK for downstream integrations
  • document privacy-safe local tracing and cover it with regression tests

v0.1.4-alpha

  • add ergonomic SDK helpers for spans, LLM calls, tool calls, and memory events
  • refactor demo agents to use the higher-level instrumentation flow
  • strengthen SDK regression coverage for the new helper methods
  • document a more realistic local-first instrumentation style in the README

v0.1.3-alpha

  • highlight the first suspicious step and likely failure step directly in the local trace viewer
  • visually emphasize error events in the timeline
  • add event-type filters for narrowing noisy traces faster
  • strengthen viewer regression coverage for the new debugging controls

v0.1.2-alpha

  • expand CLI into a more usable debugging entrypoint with demo, view, diff, and explain
  • support built-in demo scenarios for minimal, divergent, and visible-failure runs
  • add regression coverage for CLI diff/explain flows
  • document the CLI workflow in the README

v0.1.1-alpha

  • add run-level metrics to the local HTML trace viewer
  • surface total latency, input/output tokens, and tool call count at the top of each run
  • add event-type counts to the failure summary panel
  • strengthen viewer regression coverage for the new stats panel

v0.1.0-alpha

  • define project positioning and architecture draft
  • add Python SDK for local JSONL trace capture
  • add minimal example agent and a divergent failure-focused example
  • add static HTML trace viewer with failure summary
  • add run divergence generator
  • add initial failure analysis / suspicious signal heuristics
  • add divergent hidden-failure and visible-failure demo scenarios
  • add automated tests (10 passed)