ai-coding-assessment-toolkit

Use only at the end of an exam task to write a short submission-oriented README that tells graders what was built, how to run it, how it was verified, what was intentionally omitted, and what would come next. Not for marketing, changelogs, or long architecture essays.

Use only after the main path exists to choose and implement 1-3 critical-path tests that prove the MVP closes the loop. Not for broad coverage goals, full E2E suites, or testing every edge case.

Use only when the user explicitly asks to write, improve, adapt, or decompose a prompt for another AI tool or IDE. This is an optional sidecar skill for prompt-authoring tasks, not part of the default exam coding path.

Use only after prompt intake to classify the task as gateway/proxy/aggregator/BFF, tool-using assistant, retrieval assistant, or workflow agent, then choose the smallest suitable implementation path. Not for framework comparison or coding.

Use only before implementation to define the minimum contract for an agent project: endpoint or interface, tool contract, input and output shapes, state flow, error handling, and assumptions. Not for abstract architecture design.

Use only when the exam project fails to start, fails a key test, or returns the wrong contract. Prioritize the fastest evidence path to restore the main loop. Not for large refactors or speculative cleanup.

Use only at the start of a timed agent-project exam to shrink the prompt into a 2-hour MVP with explicit scope, non-goals, happy path, file tree, and time budget. Not for implementation details.