docs: update meta files for Codex/Ollama provider runners

This commit is contained in:
Hibryda 2026-03-11 03:56:05 +01:00
parent 8309896e7d
commit ad7e24e40d
6 changed files with 68 additions and 5 deletions

View file

@ -11,6 +11,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
- `claude_read_skill` path traversal: added `canonicalize()` + `starts_with()` validation to prevent reading arbitrary files via crafted skill paths (lib.rs)
### Added
- **Codex provider runner**`sidecar/codex-runner.ts` wraps `@openai/codex-sdk` (dynamic import, graceful failure if not installed). Maps Codex ThreadEvents (agent_message, reasoning, command_execution, file_change, mcp_tool_call, web_search) to common AgentMessage format via `codex-messages.ts` adapter. Sandbox/approval mode mapping from BTerminal permission modes. Session resume via thread ID. `providers/codex.ts` ProviderMeta (gpt-5.4 default, hasSandbox, supportsResume). 19 adapter tests
- **Ollama provider runner**`sidecar/ollama-runner.ts` uses direct HTTP to `localhost:11434/api/chat` with NDJSON streaming (zero external dependencies). Health check before session start. Configurable host/model/num_ctx/think via providerConfig. Supports Qwen3 extended thinking. `ollama-messages.ts` adapter maps streaming chunks to AgentMessage (text, thinking, cost with token counts). `providers/ollama.ts` ProviderMeta (qwen3:8b default, modelSelection only). 11 adapter tests
- All 3 providers registered in App.svelte onMount + message-adapters.ts. `build:sidecar` builds all 3 runners
- **S-1 Phase 3: Worktree isolation per project** — per-project `useWorktrees` toggle in SettingsTab. When enabled, agents run in git worktrees at `<repo>/.claude/worktrees/<sessionId>/` via SDK `extraArgs: { worktree: sessionId }`. CWD-based worktree detection in agent-dispatcher (`detectWorktreeFromCwd()`) matches `.claude/`, `.codex/`, `.cursor/` worktree patterns on init events. Dual detection: CWD-based (primary) + tool_call-based (subagent fallback). 8 files, +125 lines, 7 new tests. 226 vitest + 42 cargo tests
- **S-2 Session Anchors** — preserves important conversation turns through context compaction chains. Auto-anchors first 3 turns with observation masking (reasoning preserved in full per research). Manual pin button on AgentPane text messages. Three anchor types: auto (re-injectable), pinned (display-only), promoted (user-promoted, re-injectable). Re-injection via `system_prompt` field. ContextTab anchor section with budget meter bar, per-anchor promote/demote/remove actions. SQLite `session_anchors` table with 5 CRUD commands. 5 new files, 7 modified. 219 vitest + 42 cargo tests
- **Configurable anchor budget scale**`AnchorBudgetScale` type with 4 presets: Small (2K), Medium (6K, default), Large (12K), Full (20K). Per-project 4-stop range slider in SettingsTab. `ProjectConfig.anchorBudgetScale` persisted in groups.json. ContextTab budget meter derives from project setting. agent-dispatcher resolves scale on auto-anchor