The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using the mmdc CLI for local diagram rendering and curl for interacting with the Kroki API to validate or export diagrams.
[EXTERNAL_DOWNLOADS]: The skill communicates with kroki.io, a well-known service for diagram rendering. It fetches rendered SVG and PNG files based on local diagram source code, which is an intended and documented behavior.
[PRIVILEGE_ESCALATION]: The skill includes scripts/puppeteer-config.json, which contains flags to disable the Puppeteer sandbox (--no-sandbox, --disable-setuid-sandbox). While this is a standard configuration requirement for running headless browsers in specific environments like Docker, it reduces the security isolation of the local rendering process.
[INDIRECT_PROMPT_INJECTION]: The 'Batch Mode (Repo Survey)' functionality instructs the agent to scan local codebase files (such as database schemas, API routes, and state machines) to generate diagrams. This automated ingestion of untrusted external content into the agent's context creates a potential surface for indirect prompt injection.
Ingestion points: Local project files, including SQL schemas, ORM models, and route definitions, scanned during the survey workflow (SKILL.md).
Boundary markers: The instructions require the agent to confirm the proposed diagram set with the user before generation, providing a human-in-the-loop validation step.
Capability inventory: The skill possesses the ability to execute shell commands (curl, mmdc), perform file system writes, and run a local Python script for URL generation.
Sanitization: There are no explicit instructions or mechanisms provided to sanitize or escape the content ingested from the codebase before it is used to generate diagram source files.

mermaid-skill