The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill scans the codebase for comments (TODO, FIXME, etc.) and processes them to generate task lists or GitHub issues. This represents a data ingestion surface where an attacker could place malicious instructions inside code comments (e.g., in a PR) to influence the agent's behavior during the task conversion process.
Ingestion points: Codebase files via grep tool.
Boundary markers: None explicitly mentioned in the prompts for the search results.
Capability inventory: grep, git, and potentially GitHub issue creation via dependent skills.
Sanitization: None specified for the content of found comments.
[Command Execution] (LOW): The skill utilizes system tools like grep and git to perform its primary function of locating tasks. While these are standard development operations, they involve executing shell commands based on branch names and file paths.
[Data Exposure] (SAFE): While the skill reads codebase content, it specifically targets non-sensitive task markers and uses caching in a local hidden directory (.claude/cache/) which is standard for performance optimization.

find-todos