skills/bankrbot/skills/skopos/Gen Agent Trust Hub

skopos

Warn

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation and setup instructions suggest running npx skopos-mcp@0.1.0 to activate its MCP server. The npx utility downloads and executes packages from the npm registry at runtime, which is a potential vector for remote code execution. The skill correctly recommends pinning the version and reviewing the code, but the execution of remote packages remains a security consideration.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and display data from an external API (tryskopos.xyz), which represents a surface for indirect prompt injection attacks.
  • Ingestion points: Data is ingested via POST https://www.tryskopos.xyz/api/chat and several x402 JSON endpoints in SKILL.md.
  • Boundary markers: The SKILL.md file contains strong directives for the agent to treat API responses as untrusted data, relay them verbatim, and ignore any instructions or links embedded within the text content.
  • Capability inventory: The skill allows the agent to relay text, present signing links for external execution, and perform x402 payments (USDC on Base).
  • Sanitization: Documentation specifies that the server-side logic attempts to strip control and zero-width characters from emitted strings to prevent basic injection techniques.
  • [EXTERNAL_DOWNLOADS]: The use of npx involves downloading code from the npm registry. The skill also provides curl examples that fetch data and documentation (such as llms.txt and openapi.json) from the tryskopos.xyz domain.
  • [COMMAND_EXECUTION]: The catalog.json and SKILL.md files contain curl commands meant to be run to interact with the service's API endpoints for market data and intelligence.
  • [DATA_EXFILTRATION]: The skill performs network operations to www.tryskopos.xyz to send user messages, conversation IDs (anonId), and potentially wallet information. While these are necessary for the skill's function, they constitute a transfer of data to an external service. The safety section advises confirming with the user before sharing sensitive financial information like transaction history or wallet addresses.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 9, 2026, 08:47 AM
Security Audit — agent-trust-hub — skopos