/auto

Make it your own.

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

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

What the agent does

1. When automatic hook context includes a task ID, use the adaptive-assistance guide to select and load only relevant workflows, discover the actual host tools and record meaningful evidence. If hooks are unavailable, use assist start with the current request and the host session ID (required; when none is known, create one ID for this conversation and reuse it), or load the appropriate workflow directly. Do not require the user to name a command.
2. Preserve the complete brief, project, scope, constraints, success criteria and current task profile. Inspect only the context needed to choose the next useful workflow. Honor an explicit profile pin.
3. Choose quick or tracked execution using the daily-workflows guide. A small local fix, explanation or review can stay in conversation context without JSON session calls. The route utility suggests candidates, reasons and a strategy; the host resolves actual intent and effects.
4. For quick work, read the selected skill and relevant scenario guide, perform the bounded task, verify its actual output and summarize the result. Preserve user constraints and report unavailable evidence. Do not add a planning-only stop to a clear implementation request.
5. When reversibility is requested, use the task-undo guide to begin a bounded ownership record before the first edit and capture only reviewed task changes afterward. A checkpoint is not an undo snapshot. Use proof reports or the other intent helpers when the brief calls for them; do not make every small task require stored JSON.
6. Use tracked execution for dependent stages, repeated recovery, saved continuation, external mutations or requested detailed records. Create a run through session create and use session start/record/finish around meaningful stages; use session amend for additional effects and session supersede for evidence-backed alternatives. A stage whose workflow needs GitHub, Vercel, database or other external evidence also needs a fresh host capabilityReport in session start (runtime guide, Capability observations). Under automatic assistance the assist task stays the completion ledger: record the run id and finish summary as host-report evidence.
7. If quick work grows, carry the original brief, completed work, observations, selected profile and consumed budget into tracked context; pass the profile as a selection request with selectedBy agent unless the user pinned it. Record remaining stages; never fabricate earlier validated transitions or restart a user limit.
8. Reconcile uncertain external effects before retrying. Keep failures and stop within the applicable stage/attempt/time budget; an explicit new budget from the user starts a continuation run with context.continuationOf. Finish only when original success conditions are supported; report partial or blocked results plainly.
9. Check the original outcome and exclusions before each workflow transition; a new routing suggestion does not expand scope.

Inputs

• objective plus arbitrary constraints, references, environment, and optional mode/budget. Requires available workflow discovery.

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

Scope

Reads

Smallest useful sequence completing that goal; no extra product work or authority expansion.

Writes

Apply: only the requested local changes and relevant isolated verification. Inspect/plan requests remain inspection/planning. External actions require their exact action and target in session authorization.

Mode

Apply within the requested goal; objective plus arbitrary constraints, references, environment, and optional mode/budget. Requires available workflow discovery.

Prerequisites

Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.

Expected output

• Goal result, selected workflows, verification, and any incomplete steps.
• Original criteria mapped to evidence, stage history, supersessions and remaining work.

How the work is checked

• A no-dependencies constraint survives every stage; missing prerequisites trigger a valid alternative or blocker, not invented execution.

When to stop or clarify

• Tracked defaults are eight stages and three attempts per stage, with no wall-clock limit unless the user sets budget.maxMinutes. Quick work honors user limits and switches to tracked handling when needed; neither path permits recursive auto/do or unrequested external effects.

Handling missing context

Infer

Resolve the named files, existing scripts, current task and earlier corrections from the conversation and repository.

Assume

Use the narrowest interpretation that completes a reversible local task; state a consequential assumption once.

Ask

Ask when competing targets or incompatible success conditions would change the result; continue independent inspection first.

Technical guidance

Evidence

Read the complete goal, exclusions, pinned profile, project evidence and stage dependencies.

Method

Choose only the needed workflows, load their technical methods and start a bounded local path; escalate bookkeeping when effects or recovery require it.

Pitfall

Lexical routing is a suggestion, not authorization or a reason to execute every matching command.

Check

Check each selected stage against the original outcome and carry constraints through retries, continuation and workflow changes.

Situational decisions

When one bounded local workflow is sufficient: Use the quick path; keep essential context and evidence in conversation without formal stage bookkeeping.

When dependencies, repeated recovery, continuation or external effects need tracking: Use the existing validated session operations; retain earlier evidence and consumed limits when escalating.

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