The Agent Skills Directory

[COMMAND_EXECUTION]: The skill workflow extensively uses shell commands including curl, grep, mkdir, tr, and sort to automate reconnaissance. These commands are used to download files, extract patterns, and perform differential testing against API endpoints.
[EXTERNAL_DOWNLOADS]: The skill is instructed to fetch HTML and JavaScript assets from arbitrary target domains (demonstrated with the placeholder https://console.target.com/). These files are saved to the local file system for parsing and secret harvesting.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external servers, creating a surface where malicious content within a JavaScript bundle could influence the agent's behavior.
Ingestion points: JavaScript bundles and HTML shell files are downloaded from target hosts and stored in a bundles/ directory (SKILL.md, Step 1).
Boundary markers: The skill does not implement boundary markers or instructions to treat the downloaded code content strictly as data, leaving the agent susceptible to embedded instructions.
Capability inventory: The skill utilizes shell execution and network access to perform its reconnaissance tasks.
Sanitization: There is no evidence of sanitization or safety checks performed on the downloaded content prior to processing it with regular expressions.

hunt-spa-api