Skip to content

RELEASES.md

Latest commit

 

History

History
48 lines (36 loc) · 4.99 KB

RELEASES.md

File metadata and controls

48 lines (36 loc) · 4.99 KB
Raw

Release Notes

v0.2.7

  • Terminal image previews now support Kitty graphics protocol with a half-block fallback for terminals without inline image support.
  • Config Web adds a health-check tab, permission preview, OAuth panel, unified add-model setup, GitHub Copilot OAuth defaults, and clearer ChatGPT subscription challenge warnings.
  • Agent runtime adds path-aware and command-aware permission policy v2, smart patch-kind check selection, untracked file review previews, and final report output/metrics improvements.
  • MCP/project support improves HTTP health checks, server trust gating, windowed file reads, session listing, and the project skill path migration to .agents/skills.
  • Benchmarks now include execution fixtures, login-validation demos, token-only usage metrics, and review confidence display.
  • The npm wrapper and release metadata are versioned as 0.2.7 for the matching GitHub release assets.

v0.2.6

  • Image Maker now uses image-generation models from models.json like other role routes; add them with API type openai-images.
  • Image-input routing is stricter: attached images require routeBrain planning and vision-capable text role chains, while Image Maker is reserved for generation or editing requests.
  • Config Web combines catalog, saved-provider, provider /models, and manual model setup into one add-model flow, including Anthropic-compatible model listing.
  • GitHub Copilot OAuth now defaults to public github.com; GitHub Enterprise domain input is hidden behind an explicit option.
  • ChatGPT subscription model tests now detect browser or Cloudflare challenges from chatgpt.com and report a targeted OAuth-subscription warning instead of dumping the HTML challenge; the OAuth panel now marks ChatGPT subscription OAuth as not recommended for reliable calls and suggests ClIProxy API or another compatible proxy.
  • Provider context budgeting now omits inline image payload bytes from budget estimates while preserving metadata, reducing false context-overflow failures on image prompts.
  • The npm wrapper and release metadata are versioned as 0.2.6 for the matching GitHub release assets.

v0.2.5

  • The TUI now runs on Ink 7 and React 19, and assistant/help transcript rows render Markdown through markdansi, so headings, lists, inline code, code blocks, and tables no longer appear as raw Markdown.
  • Runtime and frontend dependencies were refreshed, including Pi packages 0.77.0, Vite 8.0.14, React Router 7.16.0, and TypeBox 1.1.39.
  • The npm wrapper and release metadata are versioned as 0.2.5 for the matching GitHub release assets.

v0.2.4

  • read_file now automatically expands tiny limit windows for large files and returns structured nextOffset metadata, reducing runaway page-by-page reads that bloat the transcript and provider context.

v0.2.3

  • Config Web UI is organized into tabs with the Models tab first. The usage tab shows token statistics by model, role, and runtime phase, including Recharts charts and click-through detail filters.
  • Authenticated subscription providers from Pi OAuth, including Claude Pro/Max, ChatGPT Plus/Pro Codex, and GitHub Copilot, can be selected from the model catalog without re-entering an API key.
  • TUI running status has its own one-second clock, so elapsed time keeps moving even when no token or tool event arrives.
  • TUI running status now has a lightweight text animation: the activity prefix cycles and the active status label highlights one character at a time.
  • TUI transcript folding is keyboard-driven with Ctrl+T, which toggles foldable transcript rows between expanded and collapsed. Tool calls stay concise by default and move args/result details behind the folded row. Mouse capture is only used for wheel scrolling and can be disabled with BRAINCODE_TUI_MOUSE=false.
  • Provider message-size failures are surfaced as a Braincode handoff boundary. The TUI prompts /handoff so the user can continue from a compact @@session packet instead of silently compressing the active transcript.
  • The read-only evidence cache now resets cached entries and duplicate counters after write/execute tools, reducing stale duplicate-read warnings after files or command output change.
  • The npm wrapper and release metadata are versioned as 0.2.3 for the matching GitHub release assets.

v0.2.2

  • TUI transcript folding moved off row clicks and onto Ctrl+T, so transcript selection and folding no longer compete for mouse clicks.
  • TUI transcript rendering now uses an internal viewport with Up/Down when input is empty, PageUp/PageDown, Ctrl+Up/Ctrl+Down, and mouse-wheel history scrolling, so streaming output no longer forces the whole terminal scrollback to repaint. Prompt history is available with Ctrl+P/Ctrl+N.
  • BrainPet now lives in the bottom-right footer with stable low-refresh rendering by default, and reports context-aware progress with short dry asides when available. Set BRAINCODE_TUI_ANIMATIONS=true to re-enable its animation.
  • The npm wrapper and release metadata are versioned as 0.2.2 for the matching GitHub release assets.