The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process 'long source material' and article text from users to generate visual content. This untrusted data is then used to compile an image generation prompt.
Ingestion points: Source material and article bodies analyzed in Step 1 of the 'Workflow' in SKILL.md.
Boundary markers: The skill relies on summarization but does not specify the use of clear delimiters (such as triple-backticks or specific XML-like tags) to isolate untrusted content, which may allow embedded malicious text to influence the agent's output.
Capability inventory: The skill has the ability to read from specific directories (../../styles/), write to the local filesystem (punk-assets/), and invoke an image-generation tool (image_gen).
Sanitization: The workflow requires the agent to derive concise fields (title, summary, subject) from the article body. While this summarization step helps sanitize the input, it may not be sufficient to prevent sophisticated prompt injection attacks designed to persist through the summarization process.
[COMMAND_EXECUTION]: The skill directs the agent to construct file paths using user-influenced variables such as {style-id} and {slug}.
Evidence: SKILL.md Step 5 and 6 use these variables to read from ../../styles/{style-id}/ and write to punk-assets/punk-cover/{slug}/. Without explicit validation of these identifiers against the provided catalogs, there is a potential for directory traversal or unauthorized file system operations if the execution environment does not adequately sanitize the inputs.

punk-cover