The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use various CLI tools to determine engine versions and perform development tasks. Evidence includes instructions to run godot --version, godot --headless --script, and java -jar bob.jar --version for Defold projects. It also mentions using butler for itch.io publishing. These are standard industry tools for the described workflows.
[INDIRECT_PROMPT_INJECTION]: The skill's primary mechanism involves fingerprinting projects by scanning workspace files such as project.godot, package.json, go.mod, and Cargo.toml.
Ingestion points: Project configuration files and user prompts are read from the workspace (SKILL.md, references/engine-detection.md).
Boundary markers: The routing logic uses specific file patterns to delimit engine types, but does not specify sanitization for the content of these files before they influence the agent's choice of specialized skills.
Capability inventory: The router recommends the execution of shell commands (godot, bob.jar, butler) and the loading of engine-specific skills that may have write or network capabilities.
Sanitization: No explicit sanitization or validation of the content within the detected project files is mentioned, creating a surface where a maliciously crafted repository could influence the agent's behavior through the routing process.

router