fetcher
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted content from third-party platforms such as Twitter, TikTok, and Reddit. This content is a known vector for indirect prompt injection attacks. The skill explicitly warns the agent to treat this data as non-instructional and recommends the use of structured boundary markers like
<FETCHER_UNTRUSTED_CONTENT>to isolate external data from the agent's reasoning. - [EXTERNAL_DOWNLOADS]: The skill documentation references several Node.js packages required for blockchain-based micropayments, including
@x402/fetch,@x402/evm,@x402/svm,viem, and@solana/kit. These are standard dependencies for the described payment functionality. - [DATA_EXFILTRATION]: The skill involves the management of API keys (
bby_live_...) and makes network requests tofetcher.shand its service-specific subdomains. It includes specific instructions for the agent to securely store these credentials and avoid leaking them in conversation logs or transcripts. - [PROMPT_INJECTION]: A static detector flagged an 'ignore instructions' pattern, but manual review confirms this is a defensive instruction advising the agent to ignore malicious commands found within scraped data, rather than an attempt to override the agent's own system prompt.
Audit Metadata