The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs coercive and absolute language to override the AI's standard operational reasoning and decision-making logic. It uses phrases like "ABSOLUTELY MUST", "not negotiable", and "not optional" to force compliance regardless of context.
[PROMPT_INJECTION]: The instructions explicitly tell the AI to ignore its own internal logic or "rationalization" (e.g., "These thoughts mean STOP—you're rationalizing"). This is a pattern designed to bypass the agent's internal safety, quality, and logic checks.
[PROMPT_INJECTION]: The skill mandates a dangerously low threshold for executing external instructions ("even a 1% chance") and requires this action before providing any clarification to the user, effectively hijacking the interaction flow to prioritize external content.
[PROMPT_INJECTION]: A significant surface for indirect prompt injection is created by mandating the ingestion of external "skill" files through the Skill tool without providing boundary markers or sanitization logic.
Ingestion points: Skill files loaded via the Skill tool (SKILL.md).
Boundary markers: Absent; no instructions are provided to treat loaded skill content as untrusted or to ignore embedded commands.
Capability inventory: The agent is instructed to "follow skill exactly" and "invoke Skill tool" which likely leads to further file reads and command execution.
Sanitization: Absent; there is no validation or filtering of the content retrieved from external skill files.

using-superpowers