The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes scripts/knowledge.py to extract text from help/*.md and bmod.toml files across all active skill directories. The agent is explicitly directed in references/help.md to use this aggregated content as the source for every answer and to follow the instructions found within, creating an attack surface if a malicious skill is present in the project workspace.
[DYNAMIC_EXECUTION]: The scripts/render_skill.py script employs the Jinja2 engine to generate Markdown snapshots by rendering templates stored within the skill's directory. This allows for dynamic content generation based on project configuration and customization settings. Additionally, scripts/migrate_v6.py and scripts/stamp_release.py perform dynamic loading of other local Python files using importlib.util.
[COMMAND_EXECUTION]: The skill instructions frequently require the agent to execute Python scripts via uv run and manage packages through the npx skills CLI tool. The scripts/setup.py script automates these operations, including global updates.
[EXTERNAL_DOWNLOADS]: scripts/setup.py uses urllib.request.urlopen to retrieve bmod.toml manifests from GitHub and other remote HTTPS endpoints to verify module versions. These network calls target well-known services and the author's official update channels.

bmad