The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references documentation from the cursor/plugins GitHub repository (a well-known repository for AI editor extensions) as the inspiration for its design. This reference is for informational purposes and does not involve automated downloads or execution of external code.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze and modify repository code, which constitutes an ingestion surface for untrusted data. 1. Ingestion points: The agent is instructed to read the working tree, repository instructions, and external data contracts (SKILL.md, references/task-workflows.md). 2. Boundary markers: None explicitly defined in the instructions. 3. Capability inventory: The skill allows for file modification (implementing changes), subprocess execution (running tests, using git tools), and potentially network access (reading remote facts or using authenticated CLI tools). 4. Sanitization: No explicit sanitization or filtering of codebase content is mentioned. However, the risk is inherent to the skill's purpose, and the instructions mitigate this by requiring explicit user authorization for external actions and a final diff review by the user.
[SAFE]: No suspicious network operations, credential harvesting, or unauthorized privilege escalation patterns were detected. The skill's design emphasizes user authorization for any action affecting external systems.

engineering-mode