The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides capabilities to execute and manage system subprocesses safely.
Evidence: The run and run_background capabilities in SKILL.md utilize Python's subprocess module to execute shell commands.
Security Controls: Implements a mandatory intent requirement for every execution to ensure auditability and human-readable logs.
Security Controls: Prevents system damage via a blocklist (BLOCKED_PATTERNS) that explicitly prohibits destructive commands such as recursive deletion (rm -rf /), partition formatting (mkfs), and fork bombs (:(){ :|:& };:).
Security Controls: Enforces execution limits with a default 60-second timeout (max 1 hour) to prevent denial-of-service via resource exhaustion.
Security Controls: Documentation explicitly forbids running commands as root/sudo without explicit approval and provides guidance against piping untrusted input.
Mitigation: Employs shlex.split for safe argument parsing when not running in shell mode.
[PROMPT_INJECTION]: The skill manages the risk of indirect prompt injection as it executes commands originating from other agents or task contexts.
Ingestion points: The command parameter in SKILL.md and schema.yaml serves as the entry point for external instructions.
Boundary markers: Utilizes a regex-based blocklist and mandatory timeouts as boundaries to contain process execution.
Capability inventory: Core capabilities include arbitrary subprocess execution (subprocess.run, subprocess.Popen) and process termination (os.kill).
Sanitization: Implements strict validation of input commands against the BLOCKED_PATTERNS regex and provides pre-validation through the is_blocked capability.

process-runner