The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external Excalidraw JSON files which are technically untrusted data. While these files could contain text labels intended to influence an agent's summary, the skill explicitly mandates isolating this processing within subagents and provides templates that focus on structured extraction (text labels and relationships) rather than direct execution of embedded content. This isolation strategy effectively mitigates the risk of context pollution for the main agent.
Ingestion points: Files with .excalidraw or .excalidraw.json extensions within the project workspace.
Boundary markers: The skill provides specific task templates for subagents that include explicit 'Return' constraints (e.g., 'DO NOT return raw JSON or verbose element details') and 'Approach' steps to guide the subagent's focus.
Capability inventory: The skill utilizes standard file read and write operations (Read and Write tools) and uses the platform's Task tool for subagent delegation.
Sanitization: The skill advocates for high-level summaries instead of raw data ingestion, which acts as a structural filter against direct prompt injection from the file content.

excalidraw