v0.0.9 is heavy infrastructure week on main. Still alpha — we added a license file and README polish, but have not called beta, RC, or a public v0.1.0 preview.

Shipped

Writes and reliability

  • Streaming large-file writes — .agent_write_staging/, manifest resume after truncation
  • Write integrity nudges; legacy single-mode patch tools removed
  • Compensation ledger for partial plan rollback
  • Session tool-output index; outcome scoring and trajectory logging
  • Optional self-heal lint after edits

Reasoning and routing

  • Intent classification and reasoning budget (effort + think-depth per turn)
  • External reasoning surface — think() + reason() default
  • Two-tier models — main ReAct loop vs fast sidecar JSON tasks
  • Effort learning from prior turn outcomes

Browser and documents

  • Playwright browser family + stealth init scripts
  • CAPTCHA solving (2captcha / CapSolver)
  • Document engine — compose → lint → repair → PPTX / DOCX / PDF export with quality gate

Tools and telemetry

  • self_telemetry, rename_symbol, git_worktree
  • Recipe library redesign around (intent class, tool-phase shape)
  • Settings catalog + provider presets in web UI

Project

  • MIT license file added at this milestone (later superseded by FSL-1.1-MIT in v0.0.12)
  • README screenshots for web UI and persona bootstrap

Landed next

  • Brief Task Worlds experiment (added then removed same day)
  • liminal CLI + install scripts, harness power / execution-state pass, recipe library v2

Current alpha is v0.1.2.

Current: v0.1.2 →Research ledger, identity memory, inline web chat switcher.