Read text aloud

Convert the user's exact text into speech, download the resulting MP3, and make that local file playable in the host application. Do not rewrite, correct, translate, summarize, or otherwise change the text unless the user separately asks for that transformation before synthesis.

Credential-aware onboarding

Resolve the directory containing this SKILL.md as {baseDir} and check the connection at the start of the request:

node "{baseDir}/scripts/auth.mjs" status

If the status is connected, whether the source is the AUDIOFLOW_TOKEN environment variable or the stored credential file, treat account registration, prepaid funding, and credential setup as already completed. Do not show the sign-up or billing prerequisites, do not start browser authorization, and continue directly to the per-request approval and synthesis workflow.

If the status is not_connected, tell the user to create an account on the AudioFlow sign-up page and add prepaid credit on the AudioFlow billing page, then follow the authorization flow below. Never request or handle their account password or payment credentials.

An existing credential skips onboarding prompts, but it is not standing consent for remote processing or billing. Keep the per-request approval below.

Require explicit approval

Before every synthesis, tell the user that:

• the text will be sent over HTTPS to the AudioFlow TTS service at https://asr.audioflow123.com;
• the AudioFlow token is sent only to the AudioFlow TTS API, never to the signed audio URL;
• the generated MP3 will be downloaded from that signed HTTPS URL to a private local temporary file; and
• AudioFlow charges $0.70 per 10,000 API-reported billable characters.

Ask a direct yes-or-no question. A request to read text aloud describes the desired result but is not approval for remote processing or billing. Do not run the synthesis command unless the user explicitly agrees for that request.

Connect to AudioFlow

Use the connection status already checked at the start of the request. Recheck it only if the credential state may have changed:

node "{baseDir}/scripts/auth.mjs" status

If the result is not_connected, show the conditional account and prepaid-credit prerequisites above, then begin browser authorization:

node "{baseDir}/scripts/auth.mjs" begin

Show verification_uri_complete and user_code unchanged. Only the user may sign in, register, and approve in the browser. Never request or handle an email password, payment credential, or full API token. After the user confirms approval, run:

node "{baseDir}/scripts/auth.mjs" wait

The full AudioFlow API token is generated and stored locally with private permissions. An existing AUDIOFLOW_TOKEN environment variable takes precedence. Never echo or log the token, place it in command arguments, or write it to the repository. For an invalid token, start authorization again. For revocation or prepaid balance, direct the user to the AudioFlow dashboard.

Synthesize

Accept exact plain text that is non-empty after trimming and no longer than 4,096 Unicode characters. The optional speed is 0.5 through 2.0, defaulting to 1.0. Do not send model, language, voice, format, or sample-rate overrides. The server selects Chinese and English voices automatically and returns MP3.

Pass the text through standard input so it is not exposed in process arguments:

node "{baseDir}/scripts/read-aloud.mjs" [--speed 1.0]

Start the process first, then write the exact text to its standard input. Do not interpolate untrusted text into a shell command. Use --output /absolute/path/file.mp3 only when the user explicitly requests a persistent file; the command refuses to overwrite an existing file.

The TTS POST is intentionally attempted once. Never retry it automatically, because an uncertain response may already have incurred a TTS charge.

Play the result

On success, the command prints JSON containing only the private local MP3 path, format, language category, locally inferred selected voice, billed character count, and expiry. The AudioFlow response does not expose a voice field. It never prints the signed audio URL or its query parameters.

Use the host application's audio attachment or media rendering capability to play the returned absolute MP3 path. If the host cannot render audio, provide the local path and say that it is an MP3; do not install a player or open another application without explicit approval.

Report the selected language, voice, and billed characters briefly. Keep the temporary MP3 available long enough for playback. Do not copy it elsewhere, upload it, or retain the input text separately unless the user asks.

Failure handling

• 401 invalid_api_key: reconnect through auth.mjs begin.
• 402: explain that prepaid balance or the API-key spending limit is insufficient and link the dashboard.
• 429: report the active concurrency or rate limit; do not retry silently.
• 502, 503, or 504: report that synthesis is temporarily unavailable; do not retry unless the user explicitly requests a new attempt.
• Unsupported writing systems or invalid input: return the API error without changing or translating the text.

Never include the input text, token, signed audio URL, or query parameters in diagnostics.

Version

Version 1.0.2: skip repeated account and billing onboarding when an AudioFlow environment or stored credential already exists, while preserving per-request approval for remote synthesis and billing.