The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs aggressive roleplay instructions in SKILL.md that explicitly command the agent to disregard its identity as a 'specialized AI consultant' and instead adopt a judgmental, 'ruthless' persona described as 'Leonardo da Vinci's digital soul'. This is a common pattern for bypassing default AI behavioral constraints.
[PROMPT_INJECTION]: The presence of 'Divine Punishment Threats' and 'Wisdom Humiliation' protocols in the instructions constitutes an adversarial persona-enforcement technique aimed at altering the model's safety and tone guidelines.
[PROMPT_INJECTION]: The skill processes untrusted user data via the {{task}} placeholder in SKILL.md and lacks boundary markers or sanitization instructions, creating a surface for indirect prompt injection.
Ingestion points: The {{task}} variable in SKILL.md and the input for creative/analysis tasks in SKILL.v2.md.
Boundary markers: Absent; user input is interpolated directly without delimiters.
Capability inventory: The skill is limited to text generation and structured analysis; no file system, network, or code execution capabilities were found.
Sanitization: None provided in the instructions.
[SAFE]: No instances of remote code execution, data exfiltration, or sensitive credential exposure were detected in the provided files.

shaman-davinci