The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes localized Python scripts to facilitate its operations. scripts/profile_store.py manages user-specific and brand-specific settings in standard configuration directories (e.g., ~/.config/agent-skills/), implementing a safety filter (SENSITIVE_PARTS) to prevent the storage of credentials. scripts/validate_skill.py is used to lint the skill's own metadata using yaml.safe_load(). The agent is explicitly permitted in SKILL.md to use CLI, search, and browser capabilities for project analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted content from the target project workspace, such as existing README files, project manifests, and source code scripts, to generate documentation updates. Ingestion points are defined in the SKILL.md workflow. Capability inventory includes the ability to write files and execute local commands. While the skill references an authorship integrity contract to guide prose generation, it lacks technical delimiters (e.g., XML tags or special boundary markers) to isolate untrusted project content within prompts.

lov-better-readme