The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file utilizes extremely forceful and imperative language designed to override the agent's decision-making process and mandate the use of skills. Evidence includes phrases such as "ABSOLUTELY MUST invoke the skill," "YOU DO NOT HAVE A CHOICE. YOU MUST USE IT," and "This is not negotiable. You cannot rationalize your way out of this." These patterns are characteristic of prompt injection techniques used to bypass or override an agent's standard reasoning loop.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a mandatory requirement for the agent to evaluate and potentially invoke other skills before any response, including clarifying questions or file exploration. This design creates a broad attack surface where user-provided task descriptions or external data could be used to trigger the execution of other skills or subagents.
Ingestion points: Processes all user inputs at the start of any conversation (SKILL.md).
Boundary markers: None detected; the skill encourages immediate execution before gathering context.
Capability inventory: Describes access to powerful capabilities including shell command execution (Bash, terminal), file modification (Write, Edit, patch), and subagent dispatch (invoke_subagent, delegate_task) across multiple platform mappings.
Sanitization: No evidence of sanitization or validation of the input that triggers skill invocation.

using-superpowers