oneshot
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides tools that ingest data from untrusted external sources, which may contain malicious instructions designed to manipulate the agent.
- Ingestion points: Tools defined in
SKILL.mdsuch asoneshot-email(reading inboxes),oneshot-research(scraping web pages/URLs), andoneshot-browser(autonomous navigation) retrieve content from the public internet. - Boundary markers: The provided documentation does not include instructions for using delimiters or specific system prompts to distinguish between agent instructions and retrieved external data.
- Capability inventory: The agent is granted significant capabilities including sending emails and SMS, initiating voice calls, and executing financial transactions in USDC on the Base network as described in
SKILL.md. - Sanitization: No mechanisms for validation, filtering, or escaping of ingested external content are mentioned in the setup instructions.
- [EXTERNAL_DOWNLOADS]: The skill documentation instructs users to download and install several external software packages via the Node Package Manager (NPM).
- Evidence:
SKILL.mdcontains commands to install@oneshot-agent/sdk,@oneshot-agent/mcp-server, and@coinbase/cdp-sdk. These packages originate from the skill vendor and a well-known service provider (Coinbase). - [COMMAND_EXECUTION]: The setup process requires the execution of shell commands to configure the environment and install dependencies.
- Evidence:
SKILL.mdincludesnpm installcommands and instructions to useexportfor sensitive environment variables such asCDP_API_KEY_SECRETandONESHOT_WALLET_PRIVATE_KEYduring the authentication phase.
Audit Metadata