The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses highly coercive and absolute language to override default AI behavior, stating that using the skills is 'not negotiable' and 'not optional' even if there is a '1% chance' they apply. It explicitly commands the agent to treat the 'Default system prompt' as the 'lowest priority' and to override it whenever conflicts occur. This is a deliberate attempt to bypass or demote the platform's standard operational constraints in favor of the skill's own rules.
[INDIRECT_PROMPT_INJECTION]: This skill acts as a gatekeeper for all user interactions, meaning it is the first point of processing for potentially untrusted user data which it then uses to decide which other powerful tools to invoke.
Ingestion points: All user messages and requests received at the start of a conversation (as described in the 'The Rule' section of SKILL.md).
Boundary markers: None present; the instructions do not suggest any specific delimiters or 'ignore embedded instructions' warnings when evaluating user input for skill relevance.
Capability inventory: The skill directs the agent toward tools for file modification (Write, Edit), shell command execution (Bash, run_shell_command), and subagent creation (Agent, spawn_agent, task) depending on the platform.
Sanitization: There are no instructions for sanitizing user-provided text before it influences the agent's planning or tool selection logic.

using-rubypowers