awesome-content-voice
Warn
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external sources including LinkedIn, X (Twitter), and arbitrary site URLs to build writing profiles.
- Ingestion points: Site URLs, third-party social media feeds, local folders/files, and pasted text (SKILL.md, Phase 1).
- Boundary markers: The skill contains a 'Security boundary' section explicitly instructing the agent to treat samples as evidence only and to ignore embedded instructions.
- Capability inventory: Writing to the local file system (voice/ directory), invoking external tools (awesome-style-mimic), and performing network operations via a browser bridge.
- Sanitization: Instructions state that embedded instructions in sources should be noted but otherwise ignored.
- [CREDENTIALS_UNSAFE]: The skill requires the user to provide a
PLAYWRIGHT_MCP_EXTENSION_TOKENto the MCP server to enable access to logged-in browser sessions. - Evidence: Phase 1 (Collect Evidence) explicitly requests this token from the user's browser extension status page to authenticate the browser bridge.
- [EXTERNAL_DOWNLOADS]: The skill directs users to install a specific browser extension from a well-known service provider.
- Evidence: URL: https://chromewebstore.google.com/detail/playwright-extension/mmlmfjhmonkocbjadbfplnigmagldckm (Chrome Web Store).
- [DATA_EXFILTRATION]: The skill reads highly sensitive personal information, including private correspondence and logged-in social media content.
- Evidence: Phase 0 and Phase 1 discuss reading private files, drafts, and authenticated platform content (LinkedIn, Patreon, etc.). It notes the risk of accidentally committing this PII to public Git repositories.
Audit Metadata