The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to build and execute shell commands at runtime based on service configurations. The execute_safely function in modules/execution-patterns.md uses subprocess.run with shlex.split to mitigate basic shell injection, but the tool still enables execution of arbitrary CLI tools defined in the ServiceConfig (e.g., gemini, qwen).
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (prompts and files) that are interpolated directly into command strings.
Ingestion points: The execute method and build_command function in modules/execution-patterns.md take prompt and files as arguments.
Boundary markers: None observed in the command templates (e.g., gemini -p '{prompt}' {files}).
Capability inventory: The skill uses subprocess.run to execute commands on the host system.
Sanitization: While shlex.split is used, the logic relies on simple string formatting (.format()) for prompts, which could be exploited if an external service returns malicious content that is then piped into another command.

service-registry