What problem does it solve? Before committing to a real build, teams often need to know whether an idea is technically feasible, which approach is better, or what unknowns could kill the project. This Skill structures that exploration into disposable, evidence-backed experiments instead of guesswork. ## Core Features & Use Cases - Decompose ideas into spikes: Breaks a vague idea into 2-5 independent feasibility questions framed as Given/When/Then statements, ordered by risk. - Research then build: Surfaces competing libraries and approaches, picks one, then builds a minimal runnable prototype (CLI, HTML page, or single-endpoint server) in a standalone spikes/NNN-name/ directory. - Comparison spikes: Builds competing approaches side by side (e.g., pdfjs vs camelot) and produces a head-to-head verdict table. - Verdict reporting: Each spike closes with a VALIDATED, PARTIAL, or INVALIDATED verdict plus recommendations for the real build. - Use Case: You want to know if streaming LLM tokens over websockets can stay under 100ms latency. The Skill creates spikes/001-websocket-streaming/, builds a runnable prototype, tests edge cases, and records a verdict with evidence. ## Quick Start Ask the agent to spike whether your idea is feasible, for example: "Spike whether we can parse multi-page PDFs into structured text, comparing pdfjs and camelot."
