Security risk report for @forgecat/openai_skills_speech v0.0.11
Source Integrity
Low
Profile is sourced from OpenAI's official skills repository (https://github.com/openai/skills) with clear attribution and Apache-2.0 license.
Commit hash and timestamp provided for reproducibility; no suspicious or obfuscated origins detected.
Agent Intent
Low
Content describes legitimate text-to-speech functionality with no instructions to manipulate the AI, exfiltrate data, or bypass security controls.
Documentation provides clear guardrails (e.g., 'Never modify scripts/text_to_speech.py', 'Never ask the user to paste the full key in chat'), demonstrating defensive posture rather than malicious intent.
Reference materials are educational templates for voice direction and CLI usage; no hidden instructions or prompt injection attempts detected.
Details
Evidence
Never modify `scripts/text_to_speech.py`. If something is missing, ask the user before doing anything else.
Never ask the user to paste the full key in chat. Ask them to set it locally and confirm when ready.
Do not create one-off runners (e.g., `gen_audio.py`) unless the user explicitly asks.
Permissions
Low
Profile declares no tools or file system access; it only documents how to run a bundled Python CLI script (`scripts/text_to_speech.py`) with explicit input/output paths.
Permissions are scoped to the stated purpose: text-to-speech generation via OpenAI API, with temporary files under `tmp/speech/` and outputs under `output/speech/`.
Network access is required only for live API calls and is clearly disclosed; users can test with `--dry-run` flag without network.
Details
Evidence
Use `tmp/speech/` for intermediate files (for example JSONL batches); delete when done.
Write final artifacts under `output/speech/` when working in this repo.
Dry-run (no API call; no network required; does not require the `openai` package).
MCP Risk
Low
No MCP servers are declared in the profile; the skill relies on a bundled Python CLI and the OpenAI Python SDK.
The bundled script is documented as read-only ('Never modify') and is part of the official OpenAI skills repository, reducing risk of tampering.
Network access is limited to the OpenAI Audio API endpoint and is gated by API key authentication and rate limiting (50 requests/minute).
Details
Evidence
No MCP servers defined.
Endpoint: `POST /v1/audio/speech` (OpenAI Audio API only).
Enforce 50 requests/minute. The CLI caps `--rpm` at 50.
ForgeCat