Built-in tools overview
Three tool families at a glance
| Category | Tool | One-line purpose |
|---|---|---|
| Search | unified_search | Single entry for “find anything” (knowledge base / memory / past sessions / code) |
| code_search | Fast code grep (sandbox code + KB zero/light-index code) | |
| Assets | credential | Fetch third-party credentials safely (DOORS / Jira / GitLab / Feishu tokens) |
| record_knowledge | Persist key conclusions back to the KB / long-term memory | |
| image_tools | View, edit images; extract text / tables / diagrams | |
| file_ops | Read/write sandbox files (unpack, convert, batch) | |
| Collaboration | ask_user | Ask the human when confirmation is needed (e.g., before Gerrit -2) |
| cron / event_trigger | Cron schedules + events so agents start on their own |
All tools are tenant-isolated; writes are audited; sensitive paths use
ask_userfor a second confirmation. Tools are for AI Agents—users usually don’t invoke them directly; the AI picks them as needed. This section explains what the AI can do and why it sometimes asks for confirmation.
Tools vs SKILL: responsibilities
Often confused. In one sentence:
- Tools are platform-provided, cross-scenario primitives every SKILL can call (e.g.,
unified_search,credential) - SKILL is a scenario playbook: steps, which tools to call, and what to deliver (e.g.,
code-review,patent-agent)
Rendering diagram…
Tools are low-level primitives; SKILLs are higher-level playbooks.
Custom tools
Most tools are built-in. To add your own (e.g., internal ERP or a private API client), extend via the SKILL mechanism—a manifest plus a CLI/Python entry—and the platform registers it on startup. See Skills and Parser SKILL-style extensions.