The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local shell script, scripts/multi-model/ralph-multimodel.sh, using the Bash tool. This script acts as the orchestrator for the autonomous loop and accepts user-provided task descriptions directly as arguments.
[INDIRECT_PROMPT_INJECTION]: The skill's architecture is vulnerable to indirect prompt injection due to its processing of natural language tasks in a high-capability environment.
Ingestion points: Untrusted data enters the agent's context through the primary task parameter (e.g., "<task>").
Boundary markers: No boundary markers or "ignore" instructions are present to prevent the agent from obeying malicious instructions embedded within the task input.
Capability inventory: The skill has access to powerful tools, including Bash for command execution and Write for file system modification, which are triggered based on the task analysis.
Sanitization: The documentation does not describe any sanitization, filtering, or validation steps for the input data before it influences model routing or tool usage.
[DYNAMIC_EXECUTION]: The skill facilitates autonomous code generation and subsequent execution. Using models like Codex in "full-auto" or "yolo" modes, the skill generates code (e.g., implementing APIs or fixing tests) and then executes that code via the Bash tool to verify outcomes, representing a dynamic execution pattern.

ralph-multimodel