The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes shell commands (mkdir, curl, sips, cp) using variables derived from user-provided arguments and content fetched from external websites (e.g., {output-dir}, {company-slug}, IMAGE_URL). This pattern creates a surface for command injection if these variables are not properly sanitized by the underlying execution environment.\n- [EXTERNAL_DOWNLOADS]: The skill uses curl to download image assets from arbitrary URLs discovered on external websites. While this is central to the skill's functionality, it involves fetching content from untrusted remote sources.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the web (via WebFetch) to locate image assets and metadata. An attacker-controlled website could include malicious instructions in meta tags or page content designed to influence the agent's subsequent actions.\n
Ingestion points: External websites are loaded and parsed in Step 2 of SKILL.md to extract image URLs and metadata.\n
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the processed web data.\n
Capability inventory: The skill has access to shell execution (curl, sips, mkdir, cp) and file system operations.\n
Sanitization: The workflow does not specify any validation or sanitization of the URLs or strings extracted from websites before they are passed to shell commands.

dissect-brand