The Agent Skills Directory

[SAFE]: The skill provides configuration and scripts for a developer productivity workflow. All operations are transparent and follow security best practices.
[COMMAND_EXECUTION]: The provided Node.js script executes local file system operations using readFileSync and writeFileSync to process PR descriptions. These operations are restricted to paths provided via environment variables within the controlled environment of a GitHub Action runner.
[PRIVILEGE_ESCALATION]: The GitHub Actions workflow instructions explicitly recommend least-privilege scoping (permissions: {} and specific job-level permissions) and utilize pull_request_target while specifically advising against checking out the PR head, which prevents unauthorized code execution from forks.
[DATA_EXFILTRATION]: No exfiltration patterns were detected. The script generates Markdown links to well-known AI services (Grok, Claude, ChatGPT) and uses Shields.io for badges. The prompt data transmitted is limited to the public URL of the Pull Request being reviewed.
[OBFUSCATION]: Base64 encoding is utilized solely for embedding a static SVG icon into a data URI for the Codex badge. This is a standard method for including custom logos in GitHub badges and does not hide malicious intent.
[INDIRECT_PROMPT_INJECTION]: The script processes untrusted Pull Request descriptions which could contain hidden instructions, but it incorporates protections against such attacks.
Ingestion points: The PR body is read from a file path specified in the BODY_FILE environment variable.
Boundary markers: The script uses the marker  to strictly isolate its generated section from the user-provided content.
Capability inventory: The script's capabilities are limited to text processing and file writing; it lacks network access or the ability to execute code derived from the input.
Sanitization: The input PR URL is sanitized using encodeURIComponent before being included in the generated output.

tdk-pr-review-buttons