The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs extremely imperative and coercive language designed to override the agent's standard reasoning and decision-making processes. Phrases such as "ABSOLUTELY MUST," "not negotiable," "not optional," and "You cannot rationalize your way out of this" are behavioral overrides that mirror prompt injection techniques. The "Red Flags" section specifically instructs the agent to disregard its own internal logic (e.g., "Reality: Action = task") and exploration instincts in favor of the skill's directives.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a logic that significantly increases the attack surface for indirect prompt injection by lowering the threshold for reading and executing external "skills" to "even a 1% chance."
Ingestion points: The agent is directed to ingest external skill files or tool outputs based on any user message or perceived task.
Boundary markers: No boundary markers or safety instructions are present; the skill explicitly commands the agent to "Follow skill exactly" and avoid "rationalizing" or context-gathering before invocation.
Capability inventory: The workflow requires the invocation of the Skill tool and the TodoWrite tool to manage and execute tasks derived from external content.
Sanitization: The skill lacks sanitization or validation logic for the external skills it mandates reading, instead emphasizing that "Simple things become complex" to justify their usage.

using-superpowers