plaud
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFECREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions for installing global Node.js packages and running an installer script via npx that modifies local AI client configurations. It also includes shell workflow examples using shell piping, awk, and curl.
- [CREDENTIALS_UNSAFE]: The documentation describes using the PLAUD_CLIENT_SECRET environment variable for authentication in headless environments and identifies local token storage paths at ~/.plaud/tokens.json.
- [EXTERNAL_DOWNLOADS]: The skill downloads and installs packages from the npm registry (@plaud-ai/cli, @plaud-ai/mcp) and fetches audio files via temporary signed URLs.
- [REMOTE_CODE_EXECUTION]: An automated installer (npx -y @plaud-ai/mcp@latest install) is executed to configure system-level AI client settings, which executes vendor-provided remote code during setup.
- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection when the agent processes recording transcripts and AI summaries. Ingestion points: get_transcript and get_note tools in references/mcp.md. Boundary markers: None specified in the instructions to prevent the agent from following instructions embedded in transcripts. Capability inventory: Shell execution and file writing capabilities via the Plaud CLI. Sanitization: No explicit content sanitization or filtering is mentioned.
Audit Metadata