flaxo/wynba

feat: add modes, themes, goals, compaction, and session importBrowse files at df90ba7

flaxodotdev committed a month ago · 16 changed files

Major capability expansion for coding-agent:

- Operating modes — auto, accept (confirm before file mutations),
  plan (read-only research) and manual (no tools), plus custom modes
  from ~/.coding-agent/modes/*.json via --mode and /mode.
- Themes — built-in palettes (Dracula, One Dark, Nord, …) and custom
  themes from ~/.coding-agent/themes/*.json, switchable live with
  /theme; the markdown renderer and TUI now draw through the theme.
- Goals — long-running session objectives via create_goal/get_goal/
  update_goal tools and /goal management, persisted in SQLite with
  status, elapsed time and iteration tracking.
- Context compaction — /compact folds older turns into a model-written
  structured summary (a compaction row with a cut_msg_id boundary) and
  auto-compacts when the context budget runs out.
- Session import — an `import` command that reads transcripts from
  Claude Code, Codex CLI, OpenCode, Pi and GrokBuild into the local
  session store, with a source column so re-imports refresh in place.
- Tools — todowrite/todoread (opencode-shaped todos), goal tools,
  delete_chats with confirmation, line-diff rendering for edits, and a
  read-only task tool for plan/explore sub-agents.
- Sub-agents — higher round cap plus wall-clock deadline and confirm
  timeouts so task runs always terminate with a recorded result.
- Steering & confirmations — queued mid-turn steering messages and
  yes/no tool confirmations driven from the TUI.
- DB — migrations for sessions.mode/source, todos, goals, and
  messages.reasoning_content/cut_msg_id.
- Providers — expanded registry (Bedrock, Gemini, Azure, zen and many
  more) plus reasoning_effort / thinking-level support.
- TUI fixes — welcome banner restored after reverting a session,
  arrow-key navigation into pasted text, and per-line partial
  streaming in the markdown renderer.

.gitignore

README.md

src/agent.rs

src/compact.rs

src/db.rs

src/goal.rs

src/import.rs

src/llm.rs

src/llm/providers_data.rs

src/main.rs

src/md.rs

src/mode.rs

src/theme.rs

src/tools.rs

src/tui.rs

themes/cursor.json

Major capability expansion for coding-agent: - Operating modes — auto, accept (confirm before file mutations), plan (read-only research) and manual (no tools), plus custom modes from ~/.coding-agent/modes/*.json via --mode and /mode. - Themes — built-in palettes (Dracula, One Dark, Nord, …) and custom themes from ~/.coding-agent/themes/*.json, switchable live with /theme; the markdown renderer and TUI now draw through the theme. - Goals — long-running session objectives via create_goal/get_goal/ update_goal tools and /goal management, persisted in SQLite with status, elapsed time and iteration tracking. - Context compaction — /compact folds older turns into a model-written structured summary (a compaction row with a cut_msg_id boundary) and auto-compacts when the context budget runs out. - Session import — an `import` command that reads transcripts from Claude Code, Codex CLI, OpenCode, Pi and GrokBuild into the local session store, with a source column so re-imports refresh in place. - Tools — todowrite/todoread (opencode-shaped todos), goal tools, delete_chats with confirmation, line-diff rendering for edits, and a read-only task tool for plan/explore sub-agents. - Sub-agents — higher round cap plus wall-clock deadline and confirm timeouts so task runs always terminate with a recorded result. - Steering & confirmations — queued mid-turn steering messages and yes/no tool confirmations driven from the TUI. - DB — migrations for sessions.mode/source, todos, goals, and messages.reasoning_content/cut_msg_id. - Providers — expanded registry (Bedrock, Gemini, Azure, zen and many more) plus reasoning_effort / thinking-level support. - TUI fixes — welcome banner restored after reverting a session, arrow-key navigation into pasted text, and per-line partial streaming in the markdown renderer.