The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill triggers on and processes untrusted project files (package.json, *.sh) while providing the agent with capabilities such as command execution (edgeone CLI, npm) and network operations (curl). This configuration creates a surface for indirect prompt injection, where malicious instructions embedded in the user's code could potentially influence the agent's behavior during deployment or development tasks.
Ingestion points: The skill monitors files matching *.sh and package.json to trigger its adaptation rules.
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings for the content being processed.
Capability inventory: The skill uses the edgeone CLI (for login, linking, dev server, and deployment), npm install, curl, and the present_files tool.
Sanitization: There is no mention of sanitization, escaping, or validation of the content found within the processed project files.
[COMMAND_EXECUTION]: The skill relies on the execution of several CLI tools to perform its primary functions. It provides specific instructions for running edgeone makers dev, edgeone makers deploy, edgeone login, and npm install. It specifically instructs the agent to use non-interactive flags and background execution to adapt to the constraints of a sandboxed terminal.
[CREDENTIALS_UNSAFE]: The skill provides guidance on handling authentication for the EdgeOne platform, including the use of API tokens via command-line arguments (-t <token>), environment variables (EDGEONE_PAGES_API_TOKEN), and local JSON files (.edgeone/auth.json). While these are standard management practices for the vendor's CLI, they involve the handling and storage of sensitive authentication material within the working directory.

edgeone-makers-env-adaption