The Agent Skills Directory

[SAFE]: The skill utilizes local shell commands (wc, grep) and a Python script (init-book-skill.py) to process user-provided markdown files and manage its folder structure. All operations are within the scope of its stated purpose and performed on local files provided by the user.
[COMMAND_EXECUTION]: The skill includes a helper script scripts/init-book-skill.py for initializing project structures. The script uses standard libraries for file and directory management and does not exhibit malicious behavior such as arbitrary code execution or credential harvesting.
[PROMPT_INJECTION]: The instructions contain strong procedural directives to ensure workflow consistency. These are instructional in nature and do not attempt to bypass agent safety guardrails, extract system prompts, or override core behavior.
[DATA_EXPOSURE]: Analysis of the skill body, templates, and scripts revealed no hardcoded secrets, API keys, or unauthorized attempts to access sensitive system files or environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading and processing a user-provided book.md file.
Ingestion points: The markdown file is read via subagents in Phase 3 of the workflow.
Boundary markers: The subagent prompt templates provide structured context and specific task boundaries, defining the scope of extraction.
Capability inventory: Capabilities are limited to reading file content and writing structured markdown files to a local directory.
Sanitization: The extraction process involves transforming book prose into structured templates, which mitigates the risk of direct instruction following from the source material.

skill-from-book