The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and subprocess.Popen in run_spark_agent.py and run_spark_fleet.py to execute git commands and the codex CLI. These operations are essential for managing subagent environments and executing tasks. Commands are constructed using list-based arguments to prevent shell injection.\n- [EXTERNAL_DOWNLOADS]: The skill references documentation and features of the OpenAI Codex platform in references/codex-spark.md. It requires the codex executable to be present on the system. All external references target official OpenAI services (learn.chatgpt.com).\n- [PERSISTENCE]: The bootstrap.py script installs agent profile TOML files into ~/.codex/agents or a project-level .codex/agents directory. This configuration is necessary to enable the specific Spark subagent roles defined by the skill.\n- [INDIRECT_PROMPT_INJECTION]: The subagents orchestrated by this skill process content from the local repository. Ingestion points: Subagents read repository files via the Codex CLI as configured in run_spark_agent.py and the profile TOML files. Boundary markers: Explicit instructions in run_spark_agent.py and the role-specific TOMLs command subagents to stay within assigned tasks and authority. Capability inventory: The skill uses subprocess calls for git and codex operations, writes evidence to the local filesystem, and manages candidate worktrees. Sanitization: The skill mitigates risks by using restricted sandbox modes (read-only or workspace-write), disabling network access for writer roles, and utilizing isolated candidate worktrees to ensure changes are reviewed by the main agent before integration.

gpt-engineer-spark