The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run a PowerShell script (orchestrator-poke.ps1) using the -ExecutionPolicy Bypass flag. This technique is used to override local security configurations that would otherwise prevent script execution, increasing the risk of unauthorized code running in the environment.
[PROMPT_INJECTION]: The orchestration loop presents an indirect prompt injection surface. It reads state from .codex/orchestration/queue/next_phase.json and uses the provided prompt to spawn a new subagent via codex exec. If an attacker can influence the contents of this JSON file (e.g., through a file write or pull request), they could inject malicious instructions into the subagent's context.
Ingestion points: Reads task/phase data from .codex/orchestration/queue/next_phase.json.
Boundary markers: None identified in the provided instructions to separate system instructions from phase-specific data.
Capability inventory: Uses powershell for hook execution and codex exec for spawning subagents.
Sanitization: No explicit sanitization or validation of the prompt content retrieved from the queue file is mentioned.
[EXTERNAL_DOWNLOADS]: The skill references several external GitHub repositories in its documentation, including resources from Vercel Labs. These are used for architectural reference and do not involve direct script execution or automated downloads during skill usage.

orchestrator-session