The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bash script located at ~/.claude/plugins/pua/scripts/setup-pua-loop.sh. It passes the $ARGUMENTS variable, which contains raw user input, directly to the shell command. This pattern is highly susceptible to command injection attacks.
[COMMAND_EXECUTION]: The skill is instructed to autonomously run build and test commands during each iteration cycle without requiring manual approval or user confirmation.
[PROMPT_INJECTION]: The skill contains instructions that override standard agent behavior by explicitly forbidding the use of the AskUserQuestion tool and prohibiting the agent from expressing inability to solve a task. This creates a non-interactive autonomous loop that lacks human-in-the-loop safety checkpoints.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect injection via external data.
Ingestion points: Reads local project files and git logs (SKILL.md).
Boundary markers: Absent.
Capability inventory: Executes bash scripts, build processes, and test suites.
Sanitization: No evidence of input validation or command escaping before execution.

loop