/decision-spike

Make it your own.

In Claude Code, use the slash command and add your context. In Codex, select decision-spike from the just-vibe skill picker, then send the same brief.

Version 0.11.0 also supports /jv decision-spike, /just-vibe decision-spike and /jv:decision-spike in Claude. See shortcut setup and context examples.

What the agent does

1. Name the decision the spike will inform and the uncertainty that could change it. Define a representative workload/sample and an observable acceptance/rejection criterion before building.
2. Set a time/compute/data budget, isolated artifact location and stop rule. Choose the smallest prototype or measurement that can distinguish the alternatives; avoid building production scaffolding that does not test the hypothesis.
3. In apply mode, run within the budget and record raw observations, versions and deviations from the plan. Preserve failures and resource-limit outcomes; do not change the success threshold after seeing the result.
4. Classify supported, rejected or inconclusive. Explain the implication for the original decision and the limits of the sample; prototype success does not establish production readiness.

Inputs

• uncertainty, competing hypotheses, success criteria, time/compute budget, and scratch scope.

Optional context: scope, references, constraints, successCriteria, environment, mode, budget.

Scope

Reads

Design or explicitly run a disposable experiment; no production integration disguised as a spike.

Writes

Inspect/plan: design the experiment; save requested artifacts only. Apply: build and run the disposable prototype or measurement only in an isolated artifact location outside product source, within the declared time, compute and data budget. Production integration, shared services and paid resources require their resolved target and existing session authorization.

Mode

Plan; uncertainty, competing hypotheses, success criteria, time/compute budget, and scratch scope. Running the spike uses apply mode within that budget.

Prerequisites

The decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.

Expected output

• Experiment protocol or results with hypotheses, resource cap, observations, reusable evidence and the decision implication.

How the work is checked

• The test can reject a hypothesis; reaching the budget yields an inconclusive result rather than endless expansion.

When to stop or clarify

• Paid resources require budget authorization. Keep prototype quality and production readiness clearly distinct.

Handling missing context

Infer

Recover hard constraints, the current option, adoption status and stated priorities from the brief and prior decisions.

Assume

Without a stated budget, propose a time box and stop rule before any run.

Ask

Ask only about a missing constraint or preference that could reverse the recommendation; do not demand a complete scoring questionnaire.

Technical guidance

Evidence

Identify one uncertainty, representative input, budget and rejection condition.

Method

Run the smallest experiment that separates alternatives; retain failure evidence and mark prototype shortcuts.

Pitfall

Building a polished prototype can consume the budget without testing the disputed assumption.

Check

Apply the predeclared criterion to raw observations, including an inconclusive outcome when the sample cannot distinguish options.

Situational decisions

When budget expires without separation: Preserve measurements and report inconclusive, including what a larger experiment would need.

The coding agent follows this workflow using its available tools. Installation does not grant service access or guarantee an outcome. Read the compatibility notes.

Keep exploring