The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/process-manager.ts utility performs arbitrary shell command execution through sh -c. Specifically, the startProcess function uses a template literal to construct a background command (nohup ${proc.command}), which allows for command injection if the configuration file contains unsanitized input.- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing configuration data that may be controlled by external actors:
Ingestion points: The skill reads and executes process definitions from .claude/process-config.json located at the project root.
Boundary markers: Absent; the JSON content is parsed and used directly to construct shell commands without delimiters or safety instructions.
Capability inventory: The script has broad permissions (--allow-run, --allow-net) and can invoke sh, kill, lsof, and ps, as well as perform network fetch() requests.
Sanitization: No validation, escaping, or sanitization is performed on the command or url fields retrieved from the configuration file before use in shell execution or network operations.

process-manager