The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by scanning the codebase for API route definitions (e.g., Next.js routes, Django urlpatterns). This creates a vulnerability surface where malicious code or instructions embedded in the file structure could influence the agent's behavior during the scanning process.
Ingestion points: Codebase file discovery for route definitions (SKILL.md).
Boundary markers: None present; the agent is instructed to build a list directly from discovered patterns.
Capability inventory: Execution of curl commands and starting a local development server (SKILL.md).
Sanitization: None present; the skill treats discovered route patterns as trusted endpoint paths.
[COMMAND_EXECUTION]: The skill describes starting a development server and executing curl commands. These actions are standard for development workflows and are used here to facilitate local health checks.
[EXTERNAL_DOWNLOADS]: The skill uses curl to make network requests, but the target is strictly limited to localhost. This is a whitelisted environment for local testing and does not represent an exfiltration or remote code execution risk.

api-smoke-testing