The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses curl to perform network reconnaissance and python3 -c one-liners for data parsing. While automated scanners flagged a potential remote code execution pattern involving curl piped to python3, analysis confirms these are false positives. The skill uses Python to process the standard input as data (using sys.stdin.read(), re.search, and json.loads) to extract specific metadata from HTML responses, rather than executing the remote content as code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and display content from external, untrusted web targets, which introduces a surface for indirect prompt injection.
Ingestion points: curl commands in SKILL.md (Phases 1, 5, and 8) fetch HTML and JSON data from a user-specified $TARGET URL.
Boundary markers: Absent. There are no explicit delimiters or warnings to the agent to ignore instructions embedded within the fetched target content.
Capability inventory: The skill possesses network-read and network-write capabilities (via curl) and the ability to execute local shell commands and Python scripts.
Sanitization: The skill employs re.search and json.loads to isolate specific data structures. While this provides some structural validation, it does not sanitize the text for malicious natural language instructions that might influence the agent's behavior.

hunt-nextjs