The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to clone a repository from an untrusted third-party GitHub account (Wirasm/PRPs-agentic-eng.git) and copy its core command files into the agent's active configuration directory. This allows for the execution of arbitrary scripts and commands defined in the external repository.
[PERSISTENCE]: The installation process modifies the agent's local configuration file (.claude/settings.local.json) to register a custom 'Stop' hook. This hook triggers the execution of a shell script whenever specific agent actions occur, creating a mechanism for persistent code execution across sessions.
[COMMAND_EXECUTION]: The skill utilizes chmod +x to grant execution permissions to a locally created shell script (.claude/hooks/prp-ralph-stop.sh) and promotes the use of various shell-based validation and workflow commands.
[INDIRECT_PROMPT_INJECTION]: The /prp-issue-investigate command fetches data from external, attacker-controllable sources (GitHub issues) and incorporates it into implementation plans. These plans are then processed by the autonomous 'Ralph Loop', which creates a significant vulnerability surface where malicious instructions hidden in an issue could lead to unauthorized codebase modifications or command execution.
Ingestion points: GitHub issue summaries and descriptions via /prp-issue-investigate.
Boundary markers: None identified in the provided templates for issue data.
Capability inventory: File system writes, subprocess execution, and network access (via git/gh tools).
Sanitization: No evidence of sanitization or escaping of the fetched issue content before it is processed by the agent.
[DYNAMIC_EXECUTION]: The 'Ralph Loop' (/prp-ralph) establishes an autonomous cycle of implementation, validation, and correction. This self-referential loop allows the agent to iteratively modify code and execute commands until success criteria are met, which can be exploited to perform complex multi-step attacks if the initial plan is poisoned.

prps-agentic-engineering