The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill uses curl to fetch documentation from raw.githubusercontent.com/facebook/react. Per security guidelines, this is classified as a trusted source, downgrading the severity of the network request.
[COMMAND_EXECUTION] (SAFE): The skill provides instructions for running routine package manager tasks like npm install, npm audit, and yarn why. These are legitimate operations for the stated goal of dependency management.
[REMOTE_CODE_EXECUTION] (SAFE): Several commands use npx to run utilities like madge and react-codeshift. These are widely used developer tools, and their use here is task-appropriate.
[DATA_EXFILTRATION] (SAFE): No sensitive credentials or private files are accessed or transmitted.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill has a surface for indirect injection as it processes external content. 1. Ingestion points: Reads local src/**/*.tsx files and remote CHANGELOG.md files. 2. Boundary markers: None identified. 3. Capability inventory: Includes file-write operations (fs.writeFileSync) and package installations. 4. Sanitization: No explicit validation of external content is performed before processing.

dependency-upgrade