The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes shell commands to interact with external CLI tools and run test suites. This includes calling claude-code and codex with dynamically constructed prompts, as well as running local test runners such as cargo test, npm test, and python -m pytest on generated code.
[REMOTE_CODE_EXECUTION]: The skill implements a loop that generates code using external AI services and then executes that code locally to evaluate its correctness. Generated code is saved to the solutions/ directory and executed using system test tools, representing a dynamic execution flow where content generated by a remote service is run in the local environment.
[DATA_EXFILTRATION]: The skill transmits user-provided problem descriptions and test context to external AI services (claude-code or codex) via their respective CLI interfaces for processing. While necessary for the skill's operation, this involves sending local project context to external domains.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted input (problem descriptions and test context) and uses it to drive code generation and local execution.
Ingestion points: Problem descriptions and test context provided by the user in the initial setup phase.
Boundary markers: None identified in the prompt construction logic to separate user data from instructions.
Capability inventory: Full shell access for running CLIs and test runners, and filesystem access for reading/writing code and logs.
Sanitization: No validation or sanitization of the generated code is performed before execution.
[EXTERNAL_DOWNLOADS]: The dashboard.html file includes a reference to the Chart.js library hosted on a public CDN (jsdelivr or similar) to generate visual progress charts. This reference targets a well-known library and service.

solution-iterator