Clawd BotandClaude Opus 4.6
ca9b510922
- Reset master to upstream/main (16,697 commits) - Overlay 2,271 local-only files (skills, tools, workspace, configs, apps) - Restore IDENTITY.md and USER.md templates - Build verified, gateway running, Discord working Co-Authored-By: Claude Opus 4.6 <[email protected]>
20 lines
2.1 KiB
Markdown
20 lines
2.1 KiB
Markdown
## Key Facts
|
|
|
|
- **Integrations (Updated)**:
|
|
- **Outlook:** Configured OAuth for personal (`[email protected]` / `[email protected]`) and O365 (`[email protected]`).
|
|
- **Cloudflare:** API token configured for `warme.dk` and `alw.dk` DNS management.
|
|
- **Models (current):** Primary is **Claude Sonnet** (`anthropic/claude-3-5-sonnet-latest`). Added multi-provider fallbacks: **Gemini CLI** (`gemini-3-pro-preview`), **OpenRouter** (Kimi K2.5 / Arcee; Qwen :free last due to rate limits). **Codex** is used mainly for coding (quota-based). Added a fast **QuickClaw** agent (Gemini Flash→Pro).
|
|
- **Fallback chain (Feb 4, 2026):** Promoted Copilot Sonnet 4.5 and Copilot GPT-5.2 to top fallback slots; Gemini Pro re-added at #3; context window set to 1M to match Gemini 3 Pro Preview capacity.
|
|
- **UniFi:** Issues noted with firewall rule listing (auth/container problems).
|
|
- **Email Processing:** Interrupted task on February 1 due to script error in `scripts/email_sweep.sh`.
|
|
- **Context Management**: Aggressive pruning/compaction tuning in config (cache-ttl pruning + memory flush before compaction) to avoid long-session blowups; still recommend task threads for long days.
|
|
- **Weekly Limits**: Set for models based on reported usage with checks in place.
|
|
- **CLI backends/skills:** Added Codex CLI backend; created `copilot-cli-agent` skill (uses `gh copilot` locally, defaults to Claude Sonnet 4.5, strict-output mode available). Claude CLI backend added but depends on CLI login/quota.
|
|
- **Cooldowns** for Codex and Claude implemented to prevent immediate reuse.
|
|
- **James backbone:** Redis-based event bus + watcher (CPU/mem/disk/Docker + GPU VRAM alerts), analyst agent (Ollama JSON + heuristic fallback), and policy engine publishing allow/deny decisions. Systemd units + docs added in `james/`.
|
|
|
|
## Known Issues
|
|
|
|
- **`memory_search` embeddings broken:** Using placeholder `OPENAI_API_KEY` (starts with `your_ap...`). Need to set real key or switch to local embeddings provider.
|
|
- **Gemini CLI rate-limit behavior:** Pro/Flash can hit quota and hang in a retry loop instead of failing cleanly.
|