The Agent Skills Directory

[SAFE]: The skill provides instructional templates for managing Excalidraw files efficiently. It does not include executable code, scripts, or external dependencies. The primary mechanism is isolation through subagent delegation, which is a recommended practice for handling high-token-count files. While the skill involves processing external data (Excalidraw JSON), the patterns provided include isolation boundaries and summarization instructions that help mitigate risks associated with untrusted content.
[INDIRECT_PROMPT_INJECTION]: The skill defines a pattern for ingesting external data from Excalidraw files, which constitutes a potential attack surface for indirect prompt injection. 1. Ingestion points: The skill instructs agents to read Excalidraw files using the Read tool (via subagents). 2. Boundary markers: The task templates include specific instructions for subagents to return text-only summaries and avoid raw JSON, providing a layer of isolation. 3. Capability inventory: The skill patterns include reading, modifying, and creating files on the filesystem. 4. Sanitization: Instructions recommend extracting only text elements and summarizing architecture/flow, which acts as a manual sanitization step performed by the LLM.

excalidraw