The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates a multi-step workflow by executing various local Python scripts using uv run and python3. These scripts (e.g., skf-scan-manifests.py, skf-atomic-write.py, skf-skill-inventory.py) are located within the project's _bmad/skf/shared/scripts/ directory and are part of the skill's core logic for manifest parsing, integration detection, and file management.
[EXTERNAL_DOWNLOADS]: The workflow utilizes npx skill-check for output validation and security scanning. This involves invoking a third-party tool from the npm registry, which is a standard practice in modern development environments but represents an external dependency.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data, including project architecture documents, source code manifests, and other skills' SKILL.md files. This creates a surface for indirect prompt injection. The skill includes several mitigation strategies, such as using specialized scripts for parsing prose (e.g., skf-comention-pairs.py), implementing word-boundary and section-filtering guards, and requiring evidence citations for all findings.
Ingestion points: architecture_doc_path (markdown), constituent skills' SKILL.md (markdown), and project manifest files (JSON/TOML/YAML).
Boundary markers: The skill instructions emphasize 'Zero hallucination' and require findings to trace to actual code with citations.
Capability inventory: The skill can execute local scripts, write files to the project directory, and execute a user-defined post-completion hook.
Sanitization: Parsing and analysis are largely delegated to subprocesses (Python scripts) rather than being performed directly on raw text within the LLM context.
[COMMAND_EXECUTION]: The skill supports an on_complete hook (configured via customize.toml) which allows the execution of arbitrary shell commands upon successful completion of the workflow. This is intended for integration tasks like catalog registration but could be misused if the configuration files were compromised.

skf-create-stack-skill