The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface as it reads local project files such as BUSINESS_CONTEXT.md and previous review files located in the reviews/ directory. There are no explicit boundary markers or sanitization steps defined in the instructions to separate this external data from the system prompt. However, its primary capability is writing a generated review document back to the local reviews/ directory, and the data source is typically user-controlled within the project repository.
[DATA_EXPOSURE]: The skill instructions involve processing sensitive business data, including revenue, growth metrics, and team performance. The analysis confirms that this data is accessed solely for generating local report files and there are no network-based exfiltration patterns or external URL references found in the skill body.
[COMMAND_EXECUTION]: The skill does not contain any shell command execution or dynamic context injection patterns. It relies entirely on standard file system read/write operations for its intended functionality.

quarterly-review