The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/probe_app_server.py utilizes the subprocess module to execute the local codex binary for schema generation and runtime discovery. The script securely passes arguments as lists, avoiding shell injection vulnerabilities, and is designed to probe the installed environment as part of the skill's stated purpose.
[EXTERNAL_DOWNLOADS]: The skill references documentation and archival README files from official OpenAI repositories on GitHub (github.com/openai) and official developer domains (developers.openai.com). These are treated as trusted sources for technical references and do not represent a security risk.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes JSON metadata from the local Codex service and technical documentation from external sources.
Ingestion points: Untrusted data may enter the agent's context through the output of scripts/probe_app_server.py or remote README content fetched from GitHub.
Boundary markers: The instructions lack explicit delimiters to isolate ingested technical metadata from agent instructions.
Capability inventory: The skill enables local command execution via the codex CLI and interaction with the App Server's JSON-RPC interface.
Sanitization: No explicit sanitization or filtering is performed on the ingested JSON data, relying on the model's standard guardrails.

codex-app-server