The most expensive step in coding is not implementation — it is implementing in the wrong direction. OpenClaw's official Spike Skill treats exactly that problem: before committing to a real build, run a throwaway prototype that answers "is this idea feasible". It does not chase code quality; it chases a verdict — yes, no, or partially.
The method: a five-step validation loop
Question: collapse the vague idea into one concrete feasibility question
Research: read enough docs and source to pick a credible technical approach
Build: create the smallest runnable artifact that validates or invalidates the idea
Stress: deliberately try one edge case or failure mode to see if the prototype survives
Verdict: conclude with a three-state answer — VALIDATED, PARTIAL, or INVALIDATED
Engineering discipline: small is fast
Spike constrains the artifact shape: it defaults to .tmp/openclaw-spikes/ and prefers runnable CLIs, tiny HTML, a single endpoint, or a focused test — avoiding package sprawl, Docker, env files, app frameworks, and production cleanup. The point of these constraints is to stop a spike from quietly growing into a real project.
For multi-idea comparisons, the skill suggests splitting into 2-5 independent questions, running the riskiest one first, and keeping inputs equal while measuring the same dimensions for A/B tests — discipline that makes verdicts comparable and reproducible.
Typical use cases
Tech selection validation: "can this library do X?" — run a minimal demo to observe real behavior instead of trusting the README
A/B approach comparison: two implementation ideas, each with a minimal prototype, compared on identical inputs and metrics
Pre-build risk check: before a large project, use a spike to confirm the most uncertain technical point actually holds
How to use & download
OpenClaw users: spike is a built-in skill — trigger it in natural language ("spike this", "quick prototype", "is this possible") and artifacts land in .tmp by default, keeping the repo clean. See OpenClaw docs.
Other agent users: follows the standard SKILL.md convention; you can build the same validate-before-build workflow. Source: OpenClaw GitHub repo.
Related resources
Combine with the orchestration set: ACP Router (acp-router) handles coding-agent dispatch, TaskFlow (taskflow) orchestrates long jobs after validation, and Python debugging (python-debugpy) hunts stubborn bugs in prototypes.