The Agent Skills Directory

[PROMPT_INJECTION]: The file references/flowchart-construction.md contains a directive for the agent to treat Mermaid process diagrams as authoritative and to 'execute it exactly as written' when a diagram is presented as a procedure. While this is an instruction intended to guide agent behavior, it is contextually appropriate for a skill focused on process modeling and does not attempt to override safety filters or exfiltrate data.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow for identifying and fixing broken diagram syntax involves processing external data, which creates a surface for indirect prompt injection. By instructing the agent to follow diagrams strictly, a malicious diagram could potentially contain instructions aimed at influencing the agent's actions. This risk is minimized by the agent's restricted toolset and the lack of tools capable of executing system commands or performing arbitrary network writes.
Ingestion points: Mermaid diagram definitions provided by users for debugging or process modeling (SKILL.md, references/flowchart-construction.md).
Boundary markers: Absent; there are no specific delimiters or 'ignore' warnings for the diagrams being processed.
Capability inventory: Read, Grep, and Glob tools are allowed (SKILL.md).
Sanitization: None; the skill does not provide mechanisms to sanitize or escape instructions embedded within Mermaid syntax.
[NO_CODE]: The skill consists exclusively of Markdown reference files and Mermaid diagram definitions. It does not include any executable scripts, binaries, or software package dependencies, precluding common code-based attack vectors like remote code execution or privilege escalation.

mermaids-treasure