The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves reverse-engineering architecture by analyzing untrusted files from external repositories.
Ingestion points: references/deriving-from-code.md instructs the agent to read repository manifests such as package.json, pom.xml, go.mod, and infrastructure files like Dockerfile, docker-compose.yml, and Terraform configurations.
Boundary markers: There are no explicit instructions for the agent to treat ingested data as untrusted or to use delimiters to prevent embedded instructions from influencing behavior.
Capability inventory: The skill includes scripts/render.sh, which can execute arbitrary shell commands via docker, pnpm, and other CLI tools.
Sanitization: The instructions do not specify any sanitization or validation of the metadata extracted from these external files before processing.
[COMMAND_EXECUTION]: The skill includes a bash script designed to execute various CLI tools and Docker containers.
Evidence: scripts/render.sh executes mmdc, pnpm dlx, d2, plantuml, structurizr, and docker run to transform diagram source files into images.
[EXTERNAL_DOWNLOADS]: The rendering script is configured to download and execute code from public registries if local tools are missing.
Evidence: scripts/render.sh uses pnpm dlx @mermaid-js/mermaid-cli to fetch and execute the Mermaid CLI from the public NPM registry.
Evidence: The script uses docker run to pull and execute official images for plantuml/plantuml and structurizr/cli from Docker Hub.

c4-diagrams