The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads the full contents of files in the patent/ directory (including claims, specifications, and invention disclosures) and transmits them to an external service via the mcp__codex__codex tool. This exposes intellectual property to a third-party AI provider.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection, where malicious instructions inside patent files could influence the agent's behavior.
Ingestion points: Reads content from patent/CLAIMS.md, patent/specification/, and patent/PRIOR_ART_REPORT.md (SKILL.md).
Boundary markers: The prompt templates in Step 2 and Step 4 do not use delimiters or specific instructions to isolate the patent data from the agent's command context.
Capability inventory: The skill possesses the Bash(*), Write, and Edit tools, allowing it to execute shell commands and modify the local filesystem based on the output of the external AI.
Sanitization: No input validation or filtering is performed on the content of the patent files before they are processed.
[EXTERNAL_DOWNLOADS]: The skill's prerequisites section instructs users to install an external MCP server (codex mcp-server) from an unverified source.
[COMMAND_EXECUTION]: The skill requests broad access to the Bash tool, creating a significant attack surface. In the event of a prompt injection via a malicious patent file, the agent could be manipulated into executing harmful shell commands.

patent-review