The Agent Skills Directory

[SAFE]: The skill implements a standard documentation workflow. It reads local project files (README, package.json, marketing copy) to summarize product context and writes the results to a local file (.agents/product-marketing.md).
[DATA_EXPOSURE]: While the skill reads project files, it does so to fulfill its stated purpose of summarizing product information for the user. It does not attempt to access sensitive directories (like .ssh or .aws) or perform any network operations to send data externally.
[COMMAND_EXECUTION]: No shell commands or subprocess executions are defined in the instructions. The skill operates entirely within the agent's file system and conversational capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the codebase (README, etc.) which is a common pattern for documentation agents. While this is a theoretical attack surface, the skill has no dangerous capabilities (like network access or code execution) that could be exploited through such an injection, and the impact is limited to the content of the generated marketing document.

product-marketing