The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external configuration and CLI outputs to manage environment lifecycles, which creates a surface for indirect instructions.
Ingestion points: Reads from orca.yaml and parses outputs from provider CLIs (e.g., Vercel, Docker, SSH) to determine the state of the environment.
Boundary markers: The instructions advise the agent to sequence setup phases and seek user confirmation before executing potentially high-impact actions.
Capability inventory: The skill can write shell scripts to the local filesystem and execute commands that interact with external cloud providers.
Sanitization: Instructions specify the use of proper escaping for shell variables (e.g., \$GH_TOKEN) during script generation to prevent accidental expansion or credential leakage.
[DYNAMIC_EXECUTION]: The skill generates and executes code dynamically to automate development workflows.
Evidence: Generates temporary authentication helper scripts (e.g., /tmp/askpass.sh) at runtime to facilitate Git operations.
Evidence: Utilizes node -e for inline processing and parsing of JSON data within shell scripts to communicate state between setup phases.
[COMMAND_EXECUTION]: The skill's primary function is the orchestration of local and remote command execution.
Evidence: Scaffolds scripts for provider-specific actions such as vercel sandbox create, docker commit, and ssh management.
Evidence: Coordinates build processes inside the remote environments, including pnpm install and compiling headless applications.

orca-per-workspace-env