Skip to content

CHANGELOG.md

Latest commit

 

History

History
49 lines (36 loc) · 2.66 KB

CHANGELOG.md

File metadata and controls

49 lines (36 loc) · 2.66 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

This changelog is automatically generated by AI when releases are published.

[Unreleased]

Added

  • Telemetry Streams System: New transcript-based telemetry system for improved session tracking and metrics
    • New transcripts.db database for managing AI agent sessions and watermarks
    • Unified WatermarkStrategy abstraction supporting multiple tracking methods (byte offset, record index, timestamp, hybrid)
    • Long-lived TranscriptWorker in daemon with polling-based modification detection
    • Incremental historical transcript processing for existing sessions
    • Session/trace ID linking in metrics events for server-side correlation and analysis
    • Trailing message capture: messages after last tool call are now recorded in telemetry
    • Support for multiple agent formats: Claude Code, Cursor, Droid, GitHub Copilot
  • Enhanced Metrics Schema:
    • session_id field added to all metrics events (required, unique per agent conversation)
    • trace_id field added to checkpoint and commit events (nullable, links related operations)
    • tool_use_id field added to checkpoint events (nullable, tracks specific tool invocations)
    • Enables server-side analysis of conversation flows, tool usage patterns, and session lifecycles

Changed

  • Internal DB Migration: Existing internal.db installations automatically migrate to transcripts.db on daemon startup
    • Historical prompt records are converted to session records
    • Watermarks initialized for incremental processing of existing transcripts
    • internal_db module retained for backward compatibility but marked deprecated

Deprecated

  • internal_db module: Now deprecated in favor of transcripts module
    • All new installations use transcripts.db
    • Existing installations migrate automatically
    • Module retained only for migration purposes

Technical Details

  • Architecture: Three-component design with transcripts module, TranscriptWorker daemon task, and enhanced metrics schema
  • Processing Model: 1-second polling interval for transcript modifications, priority queue for checkpoint notifications
  • Watermarking: Per-agent configurable strategies handle diverse transcript formats (append-only JSONL, SQLite databases, etc.)
  • Migration Safety: Automatic and idempotent, preserves all historical data

For detailed design rationale and implementation notes, see docs/superpowers/specs/2026-04-29-telemetry-streams-design.md.