The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to modify file permissions using the chmod +x command on the bash scripts it creates dynamically.
[DYNAMIC_EXECUTION]: The skill generates executable shell scripts at runtime by populating a template file (template.sh) with logic and stages derived from the project's context.
[INDIRECT_PROMPT_INJECTION]: The skill reads untrusted repository files to determine the requirements and steps for the generated wizard, creating a vulnerability surface where malicious documentation or configuration could influence the agent's script generation.
Ingestion points: Reads .env, .env.example, README, docker-compose*, and .github/workflows/* to identify necessary secrets and variables.
Boundary markers: The instructions do not implement markers or specific warnings to ignore embedded instructions within these project files.
Capability inventory: The skill has the ability to write files, modify execution permissions, and generate code that interacts with local environments and GitHub's secret management CLI.
Sanitization: There is no mention of sanitizing or validating the content extracted from the repository files before using it to structure the executable script.
[CREDENTIALS_UNSAFE]: The primary function of the generated scripts is to capture sensitive data, such as API keys and CI secrets, and write them to configuration files or remote secret stores. This creates a high-privilege flow for credential handling that relies on the integrity of the generated code.

wizard