The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external markdown files (rules) and installs them into the project context (.claude/rules/). While these rules influence future agent behavior, the skill implements a manifest-based tracking system (.sd0x/install-state.json) and smart merge logic to ensure transparency and prevent unintended overrides of user-modified files.
Ingestion points: Phase 2 enumerates *.md files from a source plugin rules directory.
Boundary markers: The skill uses a Precedence: header for override files and manifest hashes for version tracking.
Capability inventory: The skill uses Write for file installation and Bash (scoped to mkdir, diff, git, and ls) for project integration.
Sanitization: The skill performs hash validation to detect drift and user modifications, though it does not filter the instructional content of the rules themselves.
[SAFE]: The skill uses local configuration paths (.sd0x/) associated with the author's namespace and does not perform any network operations or execute arbitrary remote code.

install-rules