The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill integrates with well-known development tools by invoking uv and npx to run linters such as ruff, shellcheck, and biome. These tools are retrieved from established, trusted package registries (PyPI and NPM) during execution, which is standard behavior for development-oriented tooling.
[COMMAND_EXECUTION]: Several scripts (scan-scripts.py, init_skill.py) use subprocess.run or file system operations to manage project structure and run code analysis. These operations are performed using safe practices, such as passing arguments as lists to avoid shell injection and normalizing user-provided names to prevent path traversal.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze potentially untrusted skill files. This attack surface is mitigated by the use of deterministic analysis scripts (relying on regex and Abstract Syntax Tree parsing rather than LLM interpretation for critical checks) and sanitization logic in render_report.py to prevent script injection when embedding JSON data into HTML reports.
[DATA_EXPOSURE]: The skill accesses project configuration and reads file contents within the target skill directory for the purpose of analysis. This access is scoped to the tool's primary function and does not involve unauthorized exfiltration of sensitive system files or hardcoded credentials.

bmad-workflow-builder