The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input provided through the intended-behavior variable, which represents user stories or feature requests. This data is used to generate executable diagrams and simulation code, which could be exploited if the input contains malicious instructions designed to influence the code generation process.
Ingestion points: The discover-plan-context task in SKILL.md is the entry point for the $intended-behavior input.
Boundary markers: No explicit delimiters or instructions are provided to the agent to isolate the user's prose from the code generation instructions.
Capability inventory: The skill utilizes subagent delegation, browser access for testing generated HTML (build-and-exercise-plan), and repository implementation capabilities (implement-approved-contract).
Sanitization: The workflow does not include steps to sanitize or validate the user-provided input before it is interpolated into prompts or used to drive code generation.
[DYNAMIC_EXECUTION]: The core functionality of the skill involves the runtime creation and execution of code. The build-and-exercise-plan task generates a "runnable web app or HTML page" which is then executed in a browser environment. This allows for the execution of dynamically generated JavaScript and HTML.
[COMMAND_EXECUTION]: During the discovery and implementation phases, the agent is instructed to identify and run "applicable commands" from the repository. This capability allows the agent to execute shell commands found within the project's context, which could lead to unintended command execution if the repository context is manipulated.

executable-interactive-plans