The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data including challenge files, URLs, and network service banners provided during a CTF.
Ingestion points: Challenge handout artifacts, provided files, and external URLs (SKILL.md).
Boundary markers: None present to distinguish between challenge content and potential embedded instructions for the agent.
Capability inventory: Execution of file, binwalk, exiftool, checksec, strings, and curl (SKILL.md).
Sanitization: No specific sanitization or escaping is mentioned for data read from artifacts before processing.
[COMMAND_EXECUTION]: The methodology directs the agent to execute shell commands using utilities like file, exiftool, strings, checksec, and binwalk against untrusted artifacts to identify file types and security properties.
[DATA_EXFILTRATION]: The skill performs network reconnaissance using curl to fetch headers and content from external URLs to fingerprint technology stacks, which involves network operations to domains not specified in a whitelist.

ctf-methodology