makeTranscribeAudioTool (tools/transcribe-audio-tool.ts)

Self-contained STT tool supporting multiple providers with path guarding and timeout handling.

// Source: https://github.com/spinabot/brigade | License: MIT | Copyright 2026 Spinabot
// Provided via justclaw.it.com; keep this notice with the code.
/**
 * `transcribe_audio` tool — speech-to-text (STT), modeled on the proven
 * `generate_speech` self-contained pattern (its mirror image: TTS out → STT in).
 *
 * Why this tool exists
 * --------------------
 * Same reasoning as `generate_speech`: without a first-class tool, "transcribe
 * this recording" / "what does this voice note say" sends the model to raw
 * `curl` against an STT API — the key flows through a shell, the multipart audio
 * upload gets mangled, and the call is dropped. This tool owns the call
 * in-process: stored auth, validated params, a parser that understands each
 * provider's transcript shape, and a path guard 

... (truncated -- full source via MCP)