The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection in Step 10, where it aggregates drafted specification content and original user claims into a prompt for an external reviewer tool.
Ingestion points: The skill reads several user-provided or generated files from the patent/ directory (e.g., CLAIMS.md, INVENTION_DISCLOSURE.md).
Boundary markers: The prompt template in Step 10 lacks specific boundary markers or 'ignore embedded instructions' warnings when interpolating [all claims] and [all specification sections].
Capability inventory: The skill possesses high-privilege capabilities including Bash(*), Write, Edit, and the ability to invoke other Agent and Skill tools.
Sanitization: There is no evidence of sanitization or filtering of the user-provided text before it is inserted into the final review prompt.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool for project setup (Step 1) and file writing. It specifically instructs the agent to use cat <<'EOF' heredocs to handle large files, which is a defensive coding practice that prevents shell expansion of the file content, mitigating some command injection risks.

specification-writing