The Agent Skills Directory

External Data Ingestion Surface: The skill retrieves content from external sources, such as PR descriptions and upstream changelogs, to analyze version updates.
Context: This is a characteristic of indirect prompt injection surfaces, where untrusted text is introduced into the agent's workflow. However, in this case, the activity is a requirement for the skill's intended goal of auditing dependencies.
Evidence:
Ingestion points: Pull request data via gh pr view, release notes via gh release view, and registry metadata via npm view.
Boundary markers: The instructions focus on parsing and extracting structured information from these sources.
Capability inventory: The skill utilizes standard developer tools including gh, npm, grep, and node.
Sanitization: No explicit sanitization of the external text is described, which is common for descriptive analysis tasks.
Local Script Execution: The skill includes an instruction to use node -e for inspecting local configuration files.
Context: This pattern is used to programmatically parse pnpm-lock.yaml to identify dependency relationships. It is a standard method for developers to query structured data within their local environment.
Evidence: node -e "const lock = require('./pnpm-lock.yaml'); ..." in SKILL.md.

verify-dependabot-pr