Adding a Built-In Tool

此内容尚不支持你的语言。

This guide walks through adding a new built-in tool that the LLM can call during tool-loop execution.

Steps

Create a file in src/tools/functionCalls/.
Export a class that extends BaseTool.
Define the required members:
- name: tool identifier (must be unique across all registered tools)
- description: what the tool does (shown to the LLM in the context)
- category: grouping label for logging and UI
- parameters: JSON Schema describing tool arguments
- execute(): async method that runs the tool and returns a result string
Define optional members when needed:
- requiresFeatureFlag: gate the tool behind a server config toggle
- requiresPermissions: restrict to certain Discord permission levels
- requiresFollowUp: signal that the tool result needs a follow-up generation turn
Forward context.abortSignal to every HTTP call the tool makes:
- Pass signal: context.abortSignal to raw fetch calls.
- Pass externalSignal: context.abortSignal to safeDownload calls.
- Pass it through any helper option that accepts an AbortSignal.
This gives /kill true HTTP-level cancellation. Without it, the underlying request keeps running even after the turn is stopped.
The tool is auto-discovered by toolInitializer.ts at startup, so no manual registration is needed.

If the tool should be toggled by a server config field, see docs/guides/adding-feature-flag-tool.md for the full flag-mapping steps.

bun run check    # TypeScript strict mode
bun run lint     # Biome formatting

Then test the tool by prompting the bot in a way that triggers it and checking the tool-loop logs.

docs/en/architecture/pipelines/tool-loop/: how tools are dispatched and results assembled
docs/guides/adding-feature-flag-tool.md: gating a tool behind a feature flag