The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting data from a local desktop application (Codex Desktop) through tools like read_codex_thread and read_codex_request. This creates a surface where the agent could process untrusted instructions embedded in the external content.
Ingestion points: read_codex_request, read_codex_thread, list_codex_projects, and list_codex_threads in SKILL.md.
Boundary markers: The instructions do not define specific delimiters or warn the agent to ignore instructions within the retrieved content.
Capability inventory: The skill can write back to the desktop application via start_codex_thread and send_to_codex_thread.
Sanitization: No sanitization of the retrieved content is mentioned in the provided workflow.

codex-bridge