The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions as a coordinator between multiple agent sessions, creating a potential surface for indirect prompt injection across chat boundaries.
Ingestion points: The agent reads the status, run state, pull request metadata, and CI logs from other 'Codex' chats in the inventory (SKILL.md).
Boundary markers: There are no instructions for using delimiters or specialized markers to separate the untrusted data read from other chats from the agent's internal reasoning.
Capability inventory: The skill has the authority to send instructions to other chats and direct them to perform repository operations such as resolving merge conflicts, running verification scripts, and pushing code updates (SKILL.md).
Sanitization: The instructions do not specify any sanitization or validation logic for the content read from external chat states or logs before it is used to generate new instructions.
[PERSISTENCE]: The skill includes instructions for maintaining recurring activity within the environment.
Evidence: The 'Recurring orchestration' section requires the agent to 'Create or update one five-minute heartbeat that repeats this skill's sweep' (SKILL.md). This instruction establishes a persistent loop of automated activity across multiple sessions.

orchestrate