The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by instructing the agent to ingest untrusted data from the local project environment to determine its tasks.
Ingestion points: Workflow step 1 in SKILL.md directs the agent to read the README, package metadata (e.g., package.json), and landing copy from the working directory.
Boundary markers: Absent. The instructions do not define delimiters or provide warnings to disregard instructions found within these files.
Capability inventory: The skill utilizes image generation tools and includes logic for writing files to the local disk (SVG and PNG exports).
Sanitization: Absent. There is no mention of filtering or validating the content extracted from project files before using it to influence the agent's behavior.
[COMMAND_EXECUTION]: The vector.md file provides a JavaScript code snippet that performs file system operations.
File I/O: The template uses node:fs (readFileSync and writeFileSync) to read a vector file and write multiple rasterized PNG files to the local directory.
Context: While the script is presented as a user-facing tool for export, an autonomous agent may attempt to execute this logic to fulfill the workflow requirements.

logo-maker