The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes image files and JSON-based configuration files, creating a surface for indirect prompt injection where untrusted data could influence agent behavior. Ingestion points: The skill ingests image files (e.g., input.png) and follows instructional steps defined in 'references/recipes.json' and 'references/workflows.md'. Boundary markers: Absent. There are no explicit delimiters or 'ignore embedded instructions' markers used when the agent processes image metadata or external data files. Capability inventory: The skill utilizes tools with significant filesystem capabilities, including 'app.export' (file writing), 'library.import' (file reading and movement), and 'photo.delete' (file deletion). Sanitization: Absent. No evidence of input sanitization or validation for file paths or metadata content is present.
[COMMAND_EXECUTION]: The skill executes the 'lightcraft-cli' tool to perform image editing and management. These commands have broad permissions to modify the local file system and manage photo libraries, which represents a risk if the agent's instructions are subverted.
[EXTERNAL_DOWNLOADS]: Recommends the installation of additional skills via 'npx skills add' from the vendor's repository (full-aigc-skills/lightcraft-skills). It also references external documentation and download sources (e.g., SkillHub, GitHub repositories for Lightroom/Darktable docs), which serve as external dependency vectors.

lightcraft-scenario-playbooks