The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/scaffold.py script uses subprocess.run to call a generation script (generate_targets.py) located within the repository. This is an intentional part of the tool's workflow to update generated code files based on new catalog entries.
[DYNAMIC_EXECUTION]: The script executes a Python script via a path computed from the repo-root argument. While the script name is hardcoded, the root directory can be specified by the user or agent, which is a common but sensitive pattern in automation tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes external input through command-line arguments that define tool IDs and file system paths.
Ingestion points: CLI arguments --tool, --artifact-id, --root-path, --config-path, and --relative-path in scripts/scaffold.py.
Boundary markers: The SKILL.md file contains explicit safety warnings advising against the inclusion of credentials or raw evidence in the catalogs.
Capability inventory: The tool has file system write access (Path.write_text) and the ability to execute subprocesses (subprocess.run).
Sanitization: The script implements validation via the normalize_relative_path function, which prevents the use of absolute paths, home-relative shortcuts (like $HOME), and directory traversal segments (..). It also enforces lowercase alphanumeric constraints on tool and family identifiers.

blacklight-add-agent-artifact