The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides detailed instructions for the agent to use the Bash tool for project management tasks. This includes detecting lockfiles, installing missing dependencies using various package managers (npm, pnpm, yarn, bun), and executing build and test commands like npm run build and npm run test.
[PROMPT_INJECTION]: The section titled 'AUTONOMOUS AGENT MINDSET' contains strong behavioral directives intended to increase the agent's task completion rate. It instructs the agent to 'never make excuses or give up' and to 'KEEP ITERATING until the build PASSES.' While designed to improve productivity, these instructions encourage the agent to override standard caution in favor of persistent execution.
[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that involves ingesting and acting upon untrusted user project data (source code, package.json, and lockfiles) while possessing capabilities for shell execution and network access.
Ingestion points: The agent is prompted to read and analyze local project files and dependency configuration to determine implementation steps.
Boundary markers: No explicit markers are used to separate user-provided data from system instructions.
Capability inventory: The agent has access to Bash (shell execution), Write and Edit (file modification), and WebFetch (network requests).
Sanitization: The instructions do not specify validation or sanitization of the scripts or packages found in the project environment before execution.

vercel-ai-sdk