hunt-lfi
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to download the
php_filter_chain_generatortool from a GitHub repository (https://github.com/synacktiv/php_filter_chain_generator). This external utility is then used to generate exploitation chains.\n- [COMMAND_EXECUTION]: The skill makes extensive use of command-line tools for vulnerability research, includingffuf,wfuzz,dotdotpwn,curl, andpython3. These tools are used to scan targets, test for traversal, and execute payloads.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external targets, which could contain malicious instructions designed to influence the agent's behavior.\n - Ingestion points: The skill ingests data from target URLs, HTTP response bodies, JSON payloads, and local reconnaissance files like
recon/$TARGET/urls.txt.\n - Boundary markers: No explicit delimiters or isolation protocols are defined for the ingested data, though the skill encourages manual validation of results.\n
- Capability inventory: The agent has capabilities for shell command execution, network requests, and running Python scripts.\n
- Sanitization: The skill relies on a 'Validation Discipline' section to manually verify findings and confirm that content originates from the filesystem rather than reflected input.\n- [DYNAMIC_EXECUTION]: The skill involves generating PHP payloads at runtime using a Python script (
php_filter_chain_generator.py) and executing this script to create complex exploitation strings.\n- [DATA_EXFILTRATION]: The skill recommends using Out-of-Band (OOB) techniques via Burp Collaborator subdomains to confirm blind LFI. This involves the target system sending data (such as the server's IP address) to an external service for verification.
Audit Metadata