The Agent Skills Directory

[SAFE]: The skill's primary function is to read local project files—specifically CLAUDE.md, ARCHITECTURE.md, and docs/tool-routing-report.md—to generate a rule file in .claude/rules/tool-boundary.md. All file operations are restricted to the local filesystem, and there are no network requests, external downloads, or command execution patterns identified.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests contents from markdown files to create agent instructions. However, the risk is mitigated by the mandatory prerequisite that the source routing report must be marked with status: approved by the developer, ensuring a human-in-the-loop validation process.
[Ingestion points]: Reads docs/tool-routing-report.md, CLAUDE.md, and individual SKILL.md files from the filesystem.
[Boundary markers]: The generated output is structured as a standard Markdown file with clear headers and project scope declarations.
[Capability inventory]: The skill is limited to reading and writing local files and does not have the capability to execute code or access the network.
[Sanitization]: No programmatic sanitization is performed on the ingested text, relying instead on the manual approval workflow to prevent malicious instruction persistence.

boundary