The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/scan-scripts.py and scripts/prepass.py use subprocess.run to execute external analysis tools including ruff, shellcheck, and biome, as well as internal token counting utilities. The skill also supports the execution of shell commands defined by the user in the activation_steps_prepend and activation_steps_append fields of the customize.toml configuration during agent startup.
[EXTERNAL_DOWNLOADS]: The skill utilizes uv and npx to dynamically download and run packages such as tiktoken, ruff, and @biomejs/biome. These tools are used to perform static analysis and compute metrics for the agents being built or analyzed.
[DYNAMIC_EXECUTION]: The builder process involves generating agent-specific Python and shell scripts (wake.py, init-sanctum.py) from templates (assets/wake-template.py, assets/init-sanctum-template.py) and executing them to initialize agent memory and state.
[INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The skill reads and processes the contents of external agent files located in {target-agent-path} during Analyze and Edit intent processing.
- Boundary markers: There are no explicit delimiters or boundary markers implemented to isolate external agent content from the builder's internal instructions.
- Capability inventory: The skill includes capabilities for modifying the filesystem and executing shell commands via subprocess calls.
- Sanitization: The skill does not perform sanitization on the data read from external agent files before using it in analysis lenses or including it in generated reports.

bmad-agent-builder