The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from the local codebase to identify architecture components, which creates a potential surface for indirect prompt injection if malicious instructions are hidden in the analyzed source code.
Ingestion points: Codebase files accessed via Glob, Grep, and Read tools, specifically targeting files like package.json, Dockerfile, and Terraform manifests (SKILL.md, Step 1).
Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" warnings for the agent when processing discovered codebase content.
Capability inventory: The skill utilizes filesystem read access for analysis and filesystem write access to generate the final diagram file (SKILL.md, Step 6).
Sanitization: The output is restricted to structured Excalidraw JSON data, which minimizes the risk of executing injected content.
[SAFE]: The skill operates entirely within the expected context of a developer tool. Its primary function is the generation of static diagram files, and it does not perform network operations to unknown domains, execute remote code, or attempt to escalate privileges.

excalidraw