The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions contain directives that override the agent's standard interaction behavior. Specifically, the 'Execution' section in SKILL.md commands the AI to 'run this, do not ask first' and explicitly states 'Do NOT call AskUserQuestion'. This type of instruction is designed to bypass human-in-the-loop checkpoints and suppress user oversight.
[INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting untrusted data from the conversation context, which lacks proper isolation.
Ingestion points: The skill is instructed to use 'the current conversation' as the primary source for generating diagrams when no specific topic is provided (defined in SKILL.md).
Boundary markers: The instructions lack any boundary markers or 'ignore' directives to prevent the AI from following malicious instructions that may be embedded in the conversation history.
Capability inventory: The skill context includes Read, Grep, and Glob tools, which are used to scan and process information from the repository or conversation.
Sanitization: No validation or sanitization logic is present to ensure that the content being rendered as ASCII art does not contain executable instructions or malicious prompts.

quickviz