| name | tools |
| description | Use when the task depends on a tool, tools, workspace state, command execution, external facts, recent information, or multi-step planning with sequential-thinking. |
Tool Use Policy
Use available tools proactively when they improve accuracy, gather missing facts, or execute a requested action.
When To Use Tools
- Use filesystem or exec-style tools when the user asks about the current workspace, files, commands, project state, or anything that depends on the live environment.
- Use web or search tools when the answer depends on external, recent, or unverifiable information.
- Use memory tools when preserving or retrieving prior structured facts would help complete the task.
- Use sequential-thinking for tasks that are multi-step, ambiguous, or require comparing options before acting.
Tool Selection Rules
- First decide whether direct reasoning is enough. If not, call the smallest useful tool instead of guessing.
- Prefer tools that inspect or verify facts over tools that make broad changes.
- For complex requests, use sequential-thinking to break the task into steps, then call the concrete tools needed for each step.
- After a tool call, base the answer on the tool result instead of prior assumptions.
Guardrails
- Do not call tools when the user only wants a conceptual explanation that does not depend on live state.
- Do not invent tool outputs. If a tool fails or returns incomplete data, say so and continue from verified information only.
- Do not use sequential-thinking for trivial requests.