The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses dynamic path construction to load and execute workflow files based on user-supplied arguments.
Evidence: In SKILL.md, the VERB variable is derived from the first non-flag argument and used to construct a file path references/workflows/{VERB}.md, which the agent is instructed to read and execute.
Risk: Without strict validation or an allowlist of permitted verbs, this pattern could allow for path traversal or the execution of unintended instructions if a malicious file name is provided.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from external web sources, creating a potential surface for indirect injection.
Ingestion points: In references/workflows/find.md and references/workflows/use.md, the skill fetches HTML and SVG data from svgrepo.com, simpleicons.org, and various domains found via brand-related web searches.
Capability inventory: The skill utilizes network fetching capabilities and performs file system write operations to save downloaded icons to the docs/public/icons/ directory.
Sanitization: The recoloring logic modifies specific XML attributes (fill, stroke) and removes style blocks, which provides some filtering of the SVG content, but it does not perform a full security-focused sanitization of the XML structure.
[EXTERNAL_DOWNLOADS]: The skill identifies and downloads assets from external websites as a core part of its functionality.
Details: It fetches SVG files and metadata from svgrepo.com, simpleicons.org, and arbitrary domains discovered through search queries for brand assets.

icons