earnings-revenue-guidance
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHPROMPT_INJECTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- PROMPT_INJECTION (HIGH): Vulnerable to Category 8 (Indirect Prompt Injection). The skill ingests untrusted financial transcripts via the Octagon MCP (SKILL.md). It lacks boundary markers or instructions to isolate external content, allowing malicious instructions in transcripts to potentially influence the agent. The extracted data is used for financial modeling (Capability), a high-stakes decision context.\n- REMOTE_CODE_EXECUTION (MEDIUM): Installation involves executing unverified code from an untrusted source via
npx -y octagon-mcpandnpx skills addas described in the setup documentation.\n- EXTERNAL_DOWNLOADS (MEDIUM): Setup instructions recommend piping a remote script to bash (curl | bash) for Homebrew installation, which is a dangerous execution pattern from a source outside the trust scope.\n- CREDENTIALS_UNSAFE (LOW): Setup guides for Cursor and Claude Desktop advise users to pass theOCTAGON_API_KEYvia command-line environment variables, risking exposure in shell history and system logs.
Recommendations
- AI detected serious security threats
Audit Metadata