The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements an 'auto-draft' feature that ingests untrusted data from the repository to generate marketing documentation. This creates a surface where malicious instructions embedded in project files could influence the agent's output. * Ingestion points: According to Step 2 in SKILL.md, the agent reads the README, landing pages, marketing copy, about pages, meta descriptions, and package.json. * Boundary markers: The instructions do not specify any delimiters or warnings to prevent the agent from following instructions found within the ingested files. * Capability inventory: The agent possesses capabilities to read files from the workspace and write to the .agents/product-marketing.md file. * Sanitization: There is no logic to sanitize or validate the content of the files read from the codebase before they are processed by the LLM. * Mitigation: The skill prompts the user to review, correct, and fill gaps in the auto-drafted content before it is finalized, which provides a significant manual oversight mechanism.

product-marketing