git:20260711.c28c696 to git:20260805.6a63445

1 added, 1 removed. Audit A to A.

---

name: cad-spike

- description: "Time-boxed experiment to resolve a specific unknown before betting on it - falsifiable Given/When/Then criteria tested risk-first (fail fast), a clear validated | invalidated | inconclusive verdict, throwaway code. One slim SPIKE.md, not a five-artifact wrap-up"

+ description: "Spike - a time-boxed experiment resolving one unknown before you bet on it, risk-first, with a validated | invalidated | inconclusive verdict"

argument-hint: "<the question or hypothesis to resolve>"

allowed-tools:

- Read

- Write

- Edit

- Bash

- Grep

- Glob

- AskUserQuestion

---

Answer one unknown with evidence before a plan commits to it - "will this

approach work", "is this library fast enough", "does this API do what we need".

The discipline is what makes it useful: falsifiable criteria written BEFORE the

experiment, the riskiest assumption tested first so a dead end dies fast, and an

honest verdict. The code is throwaway; the SPIKE.md verdict is the deliverable.

</objective>

<execution_context>

@${CLAUDE_PLUGIN_ROOT}/cadence-core/workflows/spike.md

</execution_context>

Run the spike workflow end-to-end. Write the Given/When/Then criteria before

building anything, order them risk-first, and stop the moment a risk-first

criterion invalidates the approach. Never soften an invalidated result into

"inconclusive" to save the idea - a spike that kills a bad approach is a success.

</process>