The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, including JSON files, CSV files, and content parsed from URLs via tools like defuddle. This data is interpolated into Markdown templates using the knap render and knap batch commands.
Ingestion points: Data enters the agent context through the --data, --data-json, and piped input mechanisms described in SKILL.md.
Boundary markers: No specific delimiters or "ignore previous instruction" guards are mentioned to separate template logic from untrusted data contents.
Capability inventory: The skill allows the agent to read local files (templates/data) and write files to the disk using the -o and --output-dir flags.
Sanitization: While the skill provides a safe_name filter to sanitize data-derived filenames, it does not explicitly describe sanitization or escaping for the main template body interpolation (e.g., {{ content }}).
[EXTERNAL_DOWNLOADS]: The instructions suggest installing the knap utility via npm install -g knap. This involves downloading code from a public package registry.
[COMMAND_EXECUTION]: The skill's primary functionality relies on executing the knap CLI and potentially other tools like defuddle in the shell to perform template rendering and batch processing operations.

knap