Super Agent modes

One slash command.
A whole pipeline

You say what kind of work this is. Super Agent decides which SDLC agents run, in what order, behind which gates, and what “done” means. Same product, eleven contracts.

Type the mode. Type the ask.

A mode is your intent, not a model name. Pick one and it is authoritative for that run — no classifier quietly swaps your /dev for something slower.

super agent composer

/auto/dev/idea/debug/fastMore

/dev add a timezone field to the account settings form

›mode chip pinned to the run · command token stripped from the task text

you choose

the mode — your situation, urgency, working style

it composes

the agents, their order, parallelism and gates

then it routes

each agent to the engine and model that fits

The five you will live in

Build it, spec it, fix it, design it, or think hard about it. These five cover almost every day of product work. The pipelines below are each mode’s contract — the orchestrator still drops optional steps that don’t earn their latency on a given run.

/dev

Development

A decided, bounded change — built, reviewed, released.

For work whose requirements are already clear: change a setting, add a field, adjust behavior, make a focused UI update. Lean, not careless.

default agents

DevReviewRelease & Verify

optional · Test, when risk or repo policy calls for it

Reviewed, then released and verified against the real environment.

/idea

Specification

A rough idea becomes an approved, buildable spec.

For when the goal is real but the scope, behavior and solution are not settled yet. It explores alternatives, exposes assumptions and converges.

default agents

Discovery + UXArchitectDocumentation

optional · Security and Performance for hard constraints

Scope, behavior, acceptance criteria, decisions, risks, open questions — and no product code until you approve.

/debug

Defect

Reproduce it, root-cause it, fix it, keep it fixed.

For a reproducible defect, regression or test failure. It reproduces before it edits, names the cause instead of masking the symptom, and keeps the fix narrow.

default agents

DebuggerDevTestReviewRelease & Verify

optional · Discovery, Security or Performance by blast radius

The root cause is recorded and regression protection ships with the fix.

/design

Interface

A polished, responsive, accessible UI change — implemented.

For work where UI/UX quality is the point, and the expected outcome is shipped interface, not a mockup or a recommendation.

default agents

DiscoveryUX & AccessibilityDevReviewRelease & Verify

optional · Architect, Test and Performance

Primary, loading, empty, error, disabled, focus, keyboard and responsive states are all covered.

/deep

High-stakes

Ambiguous, cross-system work where confidence beats speed.

For the changes you would normally block out a week for. It inspects broadly, compares alternatives, documents tradeoffs and validates assumptions first.

default agents

DiscoveryArchitectDevTest + Security + PerformanceReviewRelease & Verify

optional · UX & Accessibility and Documentation

Architecture, correctness, security, performance and operational risk each get explicit consideration.

And six for the rest of reality

Emergencies, upkeep, audits, performance, releases — and one deliberate speed override for when you have already made every decision yourself.

/fast

Speed override

Already decided and scoped? Skip the review pass.

A deliberate speed override for changes you have fully specified yourself, where an independent review is overhead rather than safety.

default agents

DevRelease & Verify

optional · None — the absence is the point

Dev still self-verifies and Release still does a final sanity pass. One fewer pass, not zero.

/maintain

Upkeep

Dependencies, config, cleanup, small reliability wins.

Narrow, low-ambiguity technical upkeep that preserves externally observable behavior unless you explicitly ask to change it.

default agents

DevTestReview

optional · Security, Documentation, Release & Verify

The system stays healthy with no unintended product change.

/incident

Emergency

Contain a live outage, then restore, then investigate.

For an active production emergency — outage, attack, leaked credential, severe degradation, data-loss risk. Containment comes first.

default agents

Debugger + SecurityDevTestRelease & Verify

optional · Discovery, Performance and Review

Evidence preserved, status communicated, follow-up remediation captured. Urgency never removes an authorization boundary.

/audit

Read-only review

Prioritized, evidence-backed findings. Nothing changes.

For when you want understanding and recommendations rather than a diff. It stays read-only and separates evidence from inference.

default agents

Review + Security + PerformanceDocumentation

optional · Discovery and UX & Accessibility

Every material finding carries evidence, consequence, priority and a practical recommendation.

/optimize

Measured improvement

Move a number: latency, throughput, cost, reliability.

Measure first, find the actual bottleneck, and never trade away correctness or maintainability without saying so out loud.

default agents

PerformanceArchitectDevTestReview

optional · Discovery and Release & Verify

A comparable before/after baseline — or the evidence for why the optimization should not proceed.

/ship

Release

Built already? Validate, release, verify.

For finished implementation work. It does not expand scope — only release-blocking corrections get made.

default agents

Test + ReviewRelease & Verify

optional · Security and Performance

Required checks and configured release gates pass before deploy; the outcome is recorded.
/auto · the default

Don’t know the mode?
Don’t pick one.

/auto isn’t an eleventh contract — it’s the router. It weighs the request, resolves to one explicit mode before anything executes, tells you why it chose that one, and snapshots the decision on the run. You can override it in one click.

And when you do pick explicitly, that wins. If your request can’t safely satisfy the mode you chose, Super Agent explains the mismatch and recommends a different one — it never silently switches contracts on you.

what it weighs

exploratory, read-only, implementation-ready or release-ready?

is production actively on fire right now?

ambiguity, system breadth, reversibility, blast radius

is this a defect or an intended behavior change?

security, privacy, data and operational risk

is a measurable optimization the actual outcome?

›resolved: /debug — reproducible failure, narrow blast radius, regression protection required

Eleven modes.
Eleven capabilities.

Modes change; the underlying SDLC capabilities stay stable. Every mode is just a different composition of these — some in sequence, some in parallel, some omitted when the risk allows it.

01

Discovery

Clarifies the problem, constraints, current behavior, affected users and success criteria.

02

Debugger

Reproduces the failure, inspects evidence, isolates the cause, proposes the smallest sound fix.

03

Architect

Defines boundaries, data flow, interfaces, tradeoffs, migrations and implementation structure.

04

UX & Accessibility

Interaction behavior, information hierarchy, states, responsive behavior, accessibility requirements.

05

Dev

Implements the approved change while preserving the behavior around it.

06

Test

Proportionate automated and manual validation, including regression coverage.

07

Security

Threats, permissions, secrets, data exposure and security-sensitive implementation details.

08

Performance

Baselines plus latency, throughput, resource use, scalability and cost.

09

Review

Independent pass on correctness, maintainability, scope, risk and test adequacy.

10

Documentation

Decisions, specifications, operational guidance and user-facing behavior, written down.

11

Release & Verify

Deploys through your configured release path and verifies the resulting environment.

Work moves between modes

A mode describes the current state, not the permanent category of a conversation. When the outcome changes, a new run starts under the right contract — linked to everything that came before it.

/idea/dev

spec approved and the work is bounded

/idea/deep

spec approved but complex or high-risk

/audit/dev or /deep

you approved the remediation

/debug/dev

the fix landed, the follow-up improvement is next

/incident/maintain, /audit or /deep

service restored, now do it properly

/design/ship

the interface is implemented and reviewed

Reaching an implementation-ready spec never auto-authorizes product code. Approving it starts a linked run with its own immutable mode snapshot and plan — and changing your composer selection only affects the next run, never an active one.

What holds in every mode

Speed modes exist because the floor never moves. These four rules apply identically to the fastest and the most paranoid contract in the catalog.

A mode is not a permission slip

Selecting a mode never grants broader authority than your actual request. Destructive, production, financial and externally visible actions keep every approval they had.

Gates cannot be modes-away

A mode may add validation. It may never bypass a required check, a repository rule, or a permission — /fast and /incident included.

Read-only means read-only

The /audit mode changes nothing. If you want the findings fixed, that is a separate run you start deliberately.

The run remembers everything

Requested mode, resolved mode, agent plan, optional-agent reasoning, gates, linked artifacts and verification status — all persisted, all snapshotted for resume.

Your agents are idle right now

Every hour your subscriptions sit unused is shipped software you didn’t get. Wake them up.