The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes bash to manage task state and execute development commands.
Evidence: Uses echo, cat, mkdir, and cp to manage IDs and progress logs in the scripts/ralph/ directory.
Evidence: Executes git add and git commit to manage version control during the execution loop.
Evidence: Invokes npm run typecheck and npm test as mandatory quality checks for each implemented task.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through its task implementation loop.
Ingestion points: The skill ingests untrusted data from user feature descriptions (Step 1) and task list descriptions (Phase 2).
Boundary markers: No explicit delimiters or instructions are used to separate task descriptions from the agent's core instructions in the handoff tool invocation.
Capability inventory: The skill has access to shell execution (bash), version control (git), task management tools, and the agent-browser for UI verification.
Sanitization: There is no evidence of sanitization or verification of the content within task descriptions before they are used to generate the goal for the handoff agent.
[REMOTE_CODE_EXECUTION]: The skill utilizes the handoff tool to dynamically execute tasks based on external instructions. While these are natural language instructions rather than binary code, they direct the agent to perform code modifications and execute shell commands on the local system.

ralph