Skip to content

CHANGELOG.md

Latest commit

 

History

History
65 lines (52 loc) · 3.22 KB

CHANGELOG.md

File metadata and controls

65 lines (52 loc) · 3.22 KB
Raw

Changelog

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

[v0.1.0] - 2026-02-28

Added

  • Hybrid agent architecture — Claude Agent SDK + Claude CLI + Codex CLI + generic CLI
    • claude-sdk agent mode via @anthropic-ai/claude-agent-sdk with structured events
    • claude agent mode via CLI spawning with stream-json output parsing
    • codex agent mode via Codex CLI with JSON output parsing
    • Generic agent mode for any CLI command accepting a prompt argument
    • --agent CLI flag to set default agent, per-task agent field in POST /api/tasks
  • Priority queue — tasks support urgent, high, normal, low priorities
  • Batch operations — POST /api/tasks/batch for submitting multiple tasks
  • Task retry — POST /api/tasks/:id/retry to requeue failed tasks
  • Task search — GET /api/tasks/search?q=keyword across prompts and output
  • Task filtering — ?status, ?limit, ?offset, ?tag query parameters
  • Budget controls — per-task maxBudget and global --total-budget limit
  • System prompt from file — --system-prompt-file flag (overrides --system-prompt)
  • Structured logging — JSON logs with debug/info/warn/error levels, --verbose/--quiet flags
  • Daily stats — GET /api/stats/daily with total, success count, cost per day
  • Budget API — GET /api/budget for remaining spend tracking
  • Performance insights — GET /api/insights with duration percentiles, success rates
  • Self-evolution system — round analysis, code review heuristics, evolution log
  • Dashboard improvements — dark/light theme, agent column, Promise.allSettled resilience
  • XSS hardening — all user-controlled values escaped in dashboard innerHTML
  • Test coverage — 71 BDD-style tests across 5 suites (AgentRunner, Scheduler, WebServer, Store, WorktreePool)
  • Pre-commit hooks — automatic tsc + test verification via .githooks/pre-commit
  • CI/CD — GitHub Actions workflow with Node 20/22 matrix, type checking, build, and test
  • Task cleanup — DELETE /api/tasks/cleanup?days=N to remove old completed tasks
  • Error endpoint — GET /api/tasks/errors for recent failures
  • Health check — GET /api/health
  • API docs — GET /api/docs

Fixed

  • Timeout race condition — handleClaudeEvent no longer overwrites timeout status with late result messages
  • Build verification — verifyBuild() now async (no longer blocks event loop for 5-15s)
  • Event cap — task events capped at 200 entries to prevent unbounded SQLite growth

Changed

  • Agent execution supports both SDK and CLI modes (hybrid architecture)
  • Dashboard uses Promise.allSettled (one failed API call no longer blanks the UI)
  • getDailyStats returns {total, success, cost, successRate} (was {count, cost, successRate})
  • Logger enhanced with level filtering and stderr routing for errors

Removed

  • Legacy Python test files and scripts

[v0.1.0-alpha] - 2026-02-27

Added

  • Multi-agent orchestration with git worktrees
  • REST API with 20+ endpoints
  • Real-time SSE events
  • Web dashboard with task submission and monitoring
  • SQLite persistence
  • Priority queue with retry logic
  • Self-evolution analysis system
  • Cost and token tracking per task