/decide

Make it your own.

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

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

What the agent does

1. Write the decision question, outcome, deadline, hard constraints and current/default option. Eliminate confirmed infeasible choices before scoring preferences; unknown feasibility is pending evidence, not a middle score.
2. Compare remaining options using evidence with dates and scope. Separate observed facts, estimates and preferences; avoid double-counting correlated criteria or presenting subjective weights as objective truth.
3. Identify the assumption capable of changing the recommendation. Vary plausible scores/weights or bounds and report whether the preferred option changes; use a bounded discriminating spike when its information is worth the delay.
4. Recommend an option or conditional reversible choice, explain consequences and rejected alternatives, and set an observable revisit trigger. Record adoption only when it occurred; a recommendation does not commit stakeholders or authorize a purchase.
5. When the user asks to save an adopted decision, use decision save with actual rationale, rejected alternatives and assumptions tied to watched files or numeric reconsideration triggers. A recommendation alone is not adoption; do not fabricate agreement.
6. For a request to revisit a saved decision, decision-revisit loads the record and history, gathers fresh attributed observations and runs decision revisit; hit thresholds are reasons to review, not permission to change architecture.
7. Do not schedule monitoring, buy services or implement the new architecture without that task scope.

Inputs

• decision, options, hard requirements, and priorities.

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

Scope

Reads

Recommend one bounded choice; no purchase, installation, or commitment.

Writes

No source changes in inspect/plan. Save only requested planning artifacts. A separately requested repair uses the relevant implementation workflow.

Mode

Plan; decision, options, hard requirements, and priorities.

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

• Decision and recommendation, feasibility/evidence table, decisive uncertainty or sensitivity, consequences, adoption status and revisit trigger.
• Adopted decision record when requested, retained history, observable triggers and attributed revisit findings.

How the work is checked

• A hard requirement outweighs a cosmetic advantage; insufficient evidence produces a conditional recommendation.

When to stop or clarify

• Ask only for preferences that can flip the outcome. Do not fabricate stakeholder agreement or imply the recommendation is already adopted.

Handling missing context

Infer

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

Assume

Compare feasible options qualitatively when weights were not supplied; make a reversible conditional recommendation when useful.

Ask

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

Technical guidance

Evidence

Gather hard constraints, current approach, realistic alternatives and evidence for the decisive unknown.

Method

Eliminate infeasible choices before comparing preferences; state which new observation would reverse the recommendation.

Pitfall

More criteria do not compensate for an unresolved mandatory requirement.

Check

Check that the chosen option satisfies every hard constraint and identify whether its advantage survives plausible uncertainty.

Situational decisions

When evidence cannot distinguish feasible options: Make a reversible conditional choice or propose a discriminating spike rather than fake certainty.

When a saved assumption may no longer hold: Evaluate fresh attributed observations and watched files against the original triggers; preserve history and separate reconsideration from implementation.

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