The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Bash script to read and update a JSON configuration file located at .claude/ru-config.json. It correctly uses jq with the --arg flag to handle user-supplied input safely, preventing command injection into the shell or the jq command itself.
[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection attack surface. It allows arbitrary text from the $ARGUMENTS variable to be saved into a configuration file. According to the skill's own documentation, these 'encouraged' items are subsequently rendered into a 'USER GUIDANCE' section that overrides agent behavior in future iterations. An attacker could potentially supply a phrase containing malicious instructions that the agent would then treat as high-priority guidance.
Ingestion points: User input via the ARGUMENTS environment variable in SKILL.md is written to .claude/ru-config.json.
Boundary markers: Absent; the script does not wrap the input in delimiters or provide warnings for the downstream process that consumes the configuration.
Capability inventory: The skill can execute Bash commands and modify project files.
Sanitization: The script uses jq to ensure valid JSON structure, but it does not perform any content-based sanitization or filtering of the instructions themselves.

encourage