The Agent Skills Directory

Command Execution: The skill uses subprocess.run in scripts/export_pdf.py to call an external web browser (Microsoft Edge or Google Chrome) for PDF conversion. While this is necessary for the skill's report-generation function, it involves executing system binaries with parameters derived from the workspace path and report title.
External Downloads: The skill downloads badge images from Microsoft's official GitHub repository (microsoft/agent-academy) using urllib.request.urlretrieve. These downloads are used for document styling and target a trusted organization's infrastructure.
Indirect Prompt Injection Surface: The skill ingests untrusted feedback data from Excel files and GitHub issues, which could contain malicious instructions designed to influence the agent during the report verification phase. Ingestion points: scripts/extract_feedback.py (Excel files) and the GitHub API (Issues). Boundary markers: The skill uses structured markdown for output but lacks strict delimiters during the ingestion of external feedback. Capability inventory: The skill possesses file writing, network download, and shell command execution capabilities. Sanitization: The build_markdown.py script implements a _clean_quote function that strips markdown headers, bolding, HTML entities, and specific text patterns from feedback entries before they are included in the report.

agent-academy-report