The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to download the installation script from moonrepo.dev, the official domain for the proto toolchain manager.
[REMOTE_CODE_EXECUTION]: Provides commands to execute the downloaded script directly via a shell pipe (bash <(curl ...)), which is a high-privilege execution pattern.
[PERSISTENCE]: Documents the proto setup and proto activate commands, which modify shell profile files (such as ~/.bashrc or ~/.zshrc) to ensure the tool and its shims are automatically loaded in future terminal sessions.
[DYNAMIC_EXECUTION]: The skill facilitates the use of WASM-based plugins which can be loaded from arbitrary remote URLs or local paths, allowing for the execution of external binary code. It also utilizes eval for shell environment activation.
[INDIRECT_PROMPT_INJECTION]: The skill processes and manages .prototools configuration files which can contain instructions for plugin sources and environment variables.
Ingestion points: Reads configuration from project-local and global .prototools files (SKILL.md, references/config.md).
Boundary markers: No specific boundary markers or sanitization logic is defined for the parsing of these configuration files within the skill instructions.
Capability inventory: The skill enables file system writes for version pinning, network requests for tool/plugin installation, and subprocess spawning for tool execution (references/commands.md).
Sanitization: The skill relies on the underlying proto binary to validate the schema and integrity of configuration files and plugin manifests.

proto