The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions to operate in "AUTONOMOUS MODE" and explicitly tells the agent "Do NOT ask questions" and to execute "without pausing for user input." These instructions attempt to override standard interactive safety protocols and human-in-the-loop confirmation for complex operations.
[COMMAND_EXECUTION]: The skill triggers multiple shell commands across different package managers including npm install, pip install, flutter pub add, and npx prisma migrate dev. These commands are executed based on the results of project file scans and user-provided arguments.
[EXTERNAL_DOWNLOADS]: The skill initiates the download of numerous third-party authentication libraries and frameworks from public registries such as npm, PyPI, and Pub.dev based on framework detection.
[DATA_EXFILTRATION]: The instructions direct the agent to scan the local environment for sensitive configuration variables such as AUTH_SECRET, NEXTAUTH_SECRET, and JWT_SECRET during the project detection phase, which creates a data exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data to determine its configuration path without adequate sanitization or boundary markers.
Ingestion points: Project manifest files (package.json, requirements.txt, Gemfile, pubspec.yaml) and user-provided $ARGUMENTS are used to determine which code to write and which packages to install.
Boundary markers: No specific delimiters or "ignore instructions" warnings are applied when the agent reads and parses project files.
Capability inventory: The skill possesses extensive capabilities including file system write access for source code and schema files, package installation, and database migration execution.
Sanitization: No sanitization or verification logic is defined for the data extracted from project files before it influences execution flow.

auth-provider