skills/mukul975/anthropic-cybersecurity-skills/performing-jwt-none-algorithm-attack/Gen Agent Trust Hub
performing-jwt-none-algorithm-attack
Fail
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from JWT tokens and remote API responses, creating a surface for potential injection if results are not handled as data only.\n
- Ingestion points: User-provided JWT strings and API endpoints via the
agent.pycommand-line interface.\n - Boundary markers: The script processes data using structured formats (JSON), though it lacks specific delimiters to isolate tool output from agent instructions.\n
- Capability inventory: The
agent.pyscript performs local file reads (for public key files) and network GET requests (for API validation).\n - Sanitization: The skill relies on standard Python libraries for JSON and Base64 processing.\n- [COMMAND_EXECUTION]: The documentation instructs the agent to execute a local Python script (
agent.py) to perform security testing operations.\n- [EXTERNAL_DOWNLOADS]: The skill references standard Python dependencies (requests,PyJWT) and educational resources from well-known security sites such as HackTricks.\n- [DATA_EXFILTRATION]: Theagent.pyscript performs network GET requests to verify token acceptance. This functionality is directed by the user's input URL for the purpose of testing authorized endpoints.
Recommendations
- Contains 2 malicious URL(s) - DO NOT USE
Audit Metadata