The Agent Skills Directory

[PROMPT_INJECTION]: The skill audits existing repository content, including workflow files and metadata. This establishes an indirect prompt injection surface where malicious data within a target repository could attempt to influence the agent's audit results.
Ingestion points: Reads existing workflow files in .github/workflows/ and repository profile data via gh api calls.
Boundary markers: The protocol describes reading and parsing files without explicitly defined delimiters for untrusted content.
Capability inventory: The skill possesses Write, Edit, and Bash capabilities for modifying repository state.
Sanitization: The skill audits content for specific security patterns (e.g., pull_request_target) to warn the user, which is a benign and intended use of the data.
[EXTERNAL_DOWNLOADS]: The skill scaffolds GitHub Actions workflows that reference external actions from trusted organizations.
Evidence: Configures actions/checkout, github/codeql-action, and actions/dependency-review-action from GitHub's official organizations.
Evidence: Configures ossf/scorecard-action from the Open Source Security Foundation (OpenSSF).
[COMMAND_EXECUTION]: The skill uses gh (GitHub CLI) and git to interact with the repository and the GitHub API.
Evidence: Uses gh api to verify and enable security features like automated security fixes, vulnerability alerts, and secret scanning.
Evidence: Uses git for repository inspection and management tasks.

oss-hygiene