The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of a local Python script (shard_tool.py) and a memory logging script (memlog.py) using the uv run command. These scripts are used for analyzing document structure, performing faithful text splits, and validating the output. This is a standard and documented use of local tools within the skill's environment.
[INDIRECT_PROMPT_INJECTION]: As the skill processes and outlines long documents which may come from external or untrusted sources, it is subject to the inherent risks of indirect prompt injection.
Ingestion points: The shard_tool.py script reads the content of user-provided markdown and text files during the outline, build, and toc subcommands (SKILL.md, scripts/shard_tool.py).
Boundary markers: The instructions do not specify the use of delimiters or specific "ignore embedded instructions" warnings when the agent is reviewing the source document or its outline.
Capability inventory: The shard_tool.py utility possesses capabilities to write files (Path.write_text), create directories (Path.mkdir), delete files (Path.unlink), and copy files (shutil.copy2).
Sanitization: The tool is designed to perform a verbatim "faithful split," meaning no sanitization or filtering of the document's content is performed.
[EXTERNAL_DOWNLOADS]: The evaluation configuration in evals/cases.json references sample documents hosted on GitHub (github.com/bmad-code-org/...). These references target the author's own repository and a well-known service, and are used solely for validation and testing purposes rather than runtime code execution.

bmad-shard-pro