The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads local patent files, including claims and invention disclosures, and transmits this content to an external AI model via the mcp__codex__codex tool. While intended for the review process, this results in the transfer of potentially sensitive intellectual property to a remote service.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of a third-party MCP server (codex mcp-server) to function. This introduces an external code dependency that is not part of the skill's own code base.
[COMMAND_EXECUTION]: The skill utilizes file system tools such as Read, Write, Edit, and Bash to aggregate patent data and generate review reports, allowing the agent to interact with and modify files in the local directory.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests content from local files and interpolates them directly into prompts sent to the reviewer model.
Ingestion points: Local files in the patent/ directory including CLAIMS.md, specification/, and INVENTION_DISCLOSURE.md.
Boundary markers: The prompt uses structural headers like CLAIMS: and SPECIFICATION SUMMARY: but lacks explicit delimiters or instructions for the model to ignore potential instructions embedded within the ingested text.
Capability inventory: Access to mcp__codex__codex for remote processing and Write/Edit for local file modifications.
Sanitization: Content is interpolated into the prompt without escaping or validation.

patent-review