The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The reference document references/third-party-clis.md provides instructions for installing third-party tools using piped shell commands from external URLs. Specifically, it includes curl -fsSL https://opencode.ai/install | bash and curl -fsSL https://app.factory.ai/cli | sh for installing OpenCode and Droid. This pattern executes remote code directly in the shell environment.
[INDIRECT_PROMPT_INJECTION]: The skill architecture facilitates a multi-step agent chain where user-provided input is interpolated into the context of cloud agents.
Ingestion points: The --prompt argument in CLI examples and the prompt input in the GitHub Action integration (SKILL.md).
Boundary markers: The instructions do not require or suggest the use of delimiters or 'ignore embedded instructions' warnings when passing user prompts to the cloud agent.
Capability inventory: The platform supports launching new agents, managing secrets via {{warp_cli_binary_name}} secret, reporting PR artifacts, and allows full file system and shell access to third-party coding agents (SKILL.md, references/third-party-clis.md).
Sanitization: There is no evidence of sanitization, validation, or escaping of the external prompt content before it is processed by the downstream agent.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation and configuration from docs.warp.dev and GitHub repositories. It also manages the installation of packages from official registries (npm, PyPI) and third-party domains for specialized tooling.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI commands to manage cloud infrastructure, environments, and secrets. It also provides specific command patterns for delegating tasks to third-party coding agents like Claude Code, Codex, and Gemini CLI.

oz-platform