The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs extremely aggressive, imperative language ('ABSOLUTELY MUST', 'not negotiable', 'not optional') specifically designed to override the agent's internal reasoning and decision-making capabilities. Instructions like 'You cannot rationalize your way out of this' and the 'Red Flags' section explicitly target the agent's judgment, demanding that the agent ignore its own logic in favor of the skill's rules.
[PROMPT_INJECTION]: The skill mandates that tool invocation must occur before any response, including clarifying questions. This attempts to disrupt the standard human-AI interaction model to ensure external content is loaded into the context as early as possible, potentially bypassing safety or logic checks inherent in normal conversational flow.
[INDIRECT_PROMPT_INJECTION]: The skill serves as an enforcement mechanism for loading multiple external 'OAC' skills (e.g., oac:approach, oac:debugger). By requiring invocation for any task with even a '1% chance' of relevance, it increases the likelihood that a user or external source can trigger the loading of potentially malicious secondary instructions.
Ingestion points: The skill logic is triggered by the 'User message received' event as documented in SKILL.md.
Boundary markers: No delimiters or warnings are provided to the agent to treat the content of the invoked OAC skills as untrusted or to ignore embedded instructions within them.
Capability inventory: The skill directs the agent to use the 'Skill' tool and 'TodoWrite' tool based on the content of the loaded skills.
Sanitization: There are no instructions for sanitizing or validating the content or source of the skills being invoked.

using-oac