The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local Python scripts and shell commands to perform directory measurements, inventory scans, and manifest updates.
Evidence: Multiple uv run and python3 calls in SKILL.md, references/execute.md, and references/select.md target project-local scripts within the {project-root}/_bmad/ or {skill-root} directories.
[DYNAMIC_EXECUTION]: The drop-roster.py script dynamically loads shared helper modules from the project's scripts directory at runtime.
Evidence: Use of importlib.util in scripts/drop-roster.py to load helper modules like skf-manifest-ops.py from a static path calculated relative to the skill's installation directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied inputs that influence subsequent file system and command operations, creating an injection surface.
Ingestion points: skill_name, version, and mode inputs are accepted via command arguments or interactive prompts in SKILL.md and references/select.md.
Boundary markers: The skill uses structured prompts and specific instructions to quote paths during subprocess assembly in references/execute.md.
Capability inventory: The agent has the capability to perform recursive directory deletion (rm -rf) and execute subprocesses (uv run) based on the processed inputs.
Sanitization: Input validation logic, such as the _plain_name check in scripts/drop-roster.py, is present to ensure that folder names do not contain path traversal characters or shell-sensitive symbols.
[SAFE]: The destructive file operations and configuration changes are intended behaviors required for the skill's primary purpose and are protected by safety guards.
Evidence: A mandatory confirmation gate is implemented in references/select.md, and the skill utilizes a specialized inventory helper to verify folder ownership and platform markers before allowing any purge operation.

skf-drop-skill