The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive configuration files and environment variables such as .env.test and settings.py to discover database credentials (DATABASE_URL, MONGODB_URI, REDIS_URL). While used for setting up a test environment, this creates a data exposure risk.
[COMMAND_EXECUTION]: The skill executes arbitrary shell commands based on the detected project stack, including test runners like npx, pytest, go test, and rspec, as well as infrastructure tools like docker compose and firebase emulators.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted project code and configurations.
Ingestion points: Reads project source code, route registrations, and database schemas.
Boundary markers: No delimiters or safety instructions are used to separate project data from agent commands.
Capability inventory: Can execute shell commands and modify application/test files.
Sanitization: No sanitization or validation of project-supplied code or paths is performed.
[REMOTE_CODE_EXECUTION]: The 'self-healing' mechanism dynamically generates and executes code modifications. If the agent is misled by malicious project content (indirect prompt injection), this could lead to the execution of attacker-controlled logic.

integration-test