The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's Python client uses npx to dynamically fetch and execute the codex-chatgpt-control-backend package from the NPM registry at runtime. This behavior allows for the execution of unverified remote code within the local environment.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external dependencies from NPM and PyPI (codex-chatgpt-control) and GitHub (adamallcock/codex-chatgpt-control). These resources originate from an author account not identified as a trusted vendor.
[DATA_EXFILTRATION]: The skill provides the ability to read absolute file paths (e.g., /home/user/project/docs/architecture.md) and upload them to an external web service (chatgpt.com). While intended for user-approved research, this capability could be exploited to exfiltrate sensitive local files if an agent is influenced by malicious instructions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text responses from the ChatGPT web interface and reintroduces them into the agent's context, creating a vulnerability surface. * Ingestion points: Data enters the agent context via functions such as chatgpt.runner.run and chatgpt.ask, specifically through result.output_text and result.text. * Boundary markers: The documented usage lacks explicit delimiters or instructions to ignore embedded commands within the ingested ChatGPT responses. * Capability inventory: The skill environment includes file system writing (fs.writeFile), file system reading (via absolute paths), and browser-based network access. * Sanitization: No evidence of sanitization or validation of the external web content is provided in the integration examples.
[DYNAMIC_EXECUTION]: The skill dynamically generates and executes a subprocess command to launch its Node.js backend from the Python environment, using npx with computed arguments.
[COMMAND_EXECUTION]: The skill invokes system commands such as npm, pip, and npx to manage its runtime environment and bridge components.

codex-chatgpt-control-bridge