The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The generate-app.sh and download-icon.sh scripts download repository metadata, configuration files, and application icons from external services including GitHub, Docker Hub, GHCR, and established icon providers (Dashboard Icons, Simple Icons, and selfh.st).
[COMMAND_EXECUTION]: The generate-app.sh script executes a python3 snippet to safely parse shell arguments from user-provided docker run commands using the standard shlex library.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data (e.g., GitHub README files, docker-compose.yml files). Maliciously crafted instructions within these external sources could theoretically influence the agent's behavior when it reviews or summarizes the generated output.
Ingestion points: Untrusted data enters via generate-app.sh when fetching content from URLs provided by the user (e.g., SKILL.md workflow steps 1 and 3).
Boundary markers: The generated output files (YAML, Markdown) do not currently use explicit boundary markers or delimiters to encapsulate untrusted content.
Capability inventory: The toolchain possesses capabilities for network operations (curl), data processing (yq, jq, python3), and filesystem writes within the specified output directory.
Sanitization: The scripts parse structured data (YAML) but do not perform sanitization on natural language content extracted from remote README files.

1panel-app-builder