Changelog

1. v0.1.0

   Added 2026-11-02

   First public build — skill convergence, MCP distribution, instruction files, rollback, an embedded terminal, provider gateway and usage accounting.

   Added

   • Skill library — duplicate detection and convergence with symlinks, junctions, hard links and copies, plus a per-skill override for the global strategy.
   • Marketplace — SkillHub (default) and ClawHub registries, with installs going through the plan drawer.
   • MCP management — one definition generated as JSON / TOML / YAML, with a diff before writing.
   • Instruction files — managed-block injection, @AGENTS.md bridging, byte-for-byte preservation outside the block.
   • Change safety — plan / apply phases, atomic writes, batch backups, one-click rollback.
   • Embedded terminal — xterm.js with a real PTY, search, export and automatic Nerd Font detection.
   • Providers and gateway — 21 built-in channels, direct / routed / aggregated modes, cross-protocol conversion, loopback only.
   • Usage — read-only parsing of five session-log sources, six ranges, grouped by agent, provider and model.
   • Agent environments — detection and install guidance for 16 agents, with every path overridable in settings.
   • Dual themes — graphite navigation on warm white, or deep graphite, with no first-paint flash.

   Known limitations

   • Provider writes cover Claude Code, Codex and OpenCode only; Gemini, CodeBuddy and Hermes are explicitly refused with the reason shown.
   • Cross-protocol conversion covers Anthropic Messages ⇄ Chat Completions. Any combination involving Responses is rejected at save time.
   • Cache-creation accounting measured zero on every local machine and is not surfaced separately.

No entries in this category yet.