The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository files and GitHub API responses, creating a surface for injection attacks.
Ingestion points: Data is read from docs/code-structure.md, ARCHITECTURE.md, package.json, and via gh api calls for PR files and issues.
Boundary markers: Absent. The skill interpolates extracted data directly into report templates and PR comments.
Capability inventory: The skill uses Bash (git, gh), Write, and Grep to interact with the filesystem and remote repositories.
Sanitization: No explicit sanitization or escaping is performed on variables (like {domain_name}) before they are used in shell commands or markdown reports.
[COMMAND_EXECUTION]: The skill relies heavily on shell command execution for its core functionality.
Evidence: Uses git to analyze repository history, commit dates, and manage branches.
Evidence: Uses the GitHub CLI (gh) to query API endpoints, list issues, and manage pull requests.
Evidence: Invokes a local shell script file-finding.sh to record audit results.

entropy-check