minmaxing

Create a taste-to-artifact visualization package for a project, task, draft SPEC, UI, backend, agent runtime, dashboard, game, or product idea without implementing it. Use when the user asks to visualize taste.md, taste.vision, what the model thinks is being built, a product experience, UI mockup, architecture schematic, operator journey, or comprehension check before execution.

Run an approval-first variant of the minmaxing workflow: research, audit, plan, estimate, draft SPEC, visualize the intended product or operator experience, then stop with WAITING_FOR_VISUAL_APPROVAL before implementation. Use when the user wants to see and approve the model's understanding before code changes, or wants to continue or revise a saved visualization workflow.

Use this as the definitive workflow command for mutating work: default Opus 4.7 high planning and review with MiniMax-M2.7-highspeed Token Plan execution, plus explicit governed model profiles such as sonnetminimax for Sonnet 4.6 judgment with MiniMax execution. It must drive to a verified result, partial result, or blocked repair path; /opusminimax is the advanced engine underneath, not a competing daily command.

Run the power-user Sonnet 4.6 judgment plus MiniMax-M2.7-highspeed Token Plan executor route. Use when the operator wants to preserve Opus quota or has exhausted Opus, while keeping the same governed /opusworkflow lifecycle and MiniMax bounded execution.

Advanced engine behind /opusworkflow for provider split, packet, repair, and benchmark work, including the explicit sonnetminimax profile for Sonnet 4.6 judgment plus MiniMax-M2.7-highspeed execution. Normal build/fix/refactor work should use /opusworkflow.

Run the optional all-Opus sibling of /opusworkflow. Use when the operator explicitly wants Opus 4.7 as planner, executor, reviewer, and final judge with default high effort and optional max effort.

Run the full minmaxing workflow end to end for one request. Use when the user wants planning, implementation, verification, and closeout to happen automatically in one command.

Diagnose Claude Code Agent View readiness safely inside the minmaxing harness. This route is static readiness and troubleshooting only; open the live Agent View TUI manually with claude agents.

Answer Claude, Claude Code, Anthropic API, skills, connectors, plugins, and Claude product questions using official current docs.

Diagnose and document native Claude Code /goal readiness safely inside the minmaxing harness. This route is static readiness and effectiveness guidance only; set native goals manually with /goal inside Claude Code.

Coordinate a governed hive of specialized agents for broad research, planning, review, or implementation work when roles, blackboard state, dissent, synthesis, and verification materially improve the outcome.

Run dense minmaxing work as a hardware-aware, main-orchestrated parallel workflow with bounded packets, explicit ownership, sync barriers, aggregation, and independent verification.

Identify the highest-leverage actions for the current product based on taste.md, taste.vision, and SPEC.md, with deepresearch-backed channel scanning, RICE+categorization scoring, auto-vs-manual classification, community targets with verified URLs, and (when research surfaces moats or blind spots) a /defineicp-style proposal to update the kernel. Use when the user invokes /leveragepath or asks "what's the highest-leverage move I can make right now?"

Digest one or more Deep Research markdown reports into a sanitized goal and taste bootstrap packet for a new or existing project. Use when the user invokes /digestaste, provides a .md deepresearch result, or asks to turn research into bootstrap text for taste.md, taste.vision, ICP, goal setup, or project direction.

Run the full minmaxing workflow from external AI research reports. Use when the task begins with Gemini Deep Research, NotebookLM, ChatGPT Deep Research, Perplexity, or similar reports that must be digested before the repo's own deepresearch and governed workflow.

QA every newly created, updated, or reused SPEC.md before implementation using current webresearch for SOTA/time-sensitive claims and an Opus 4.7 high/xhigh reviewer when runtime identity is proven.

Diagnose Claude Code native Remote Control safely inside the minmaxing harness. In this project, /remote-control is a readiness/troubleshooting skill; start the live native Remote Control server from a shell with claude remote-control.

Run the optional Claude-only Opus 4.7 planner plus Sonnet 4.6 executor workflow. Use when the user invokes /opussonnet or installed with --mode opussonnet and wants the harness without MiniMax.

Generate or edit raster image assets requested by SPEC.md using Codex subscription or ChatGPT-included image generation, not OpenAI API keys. Use when a task, SPEC.md, visualization package, README, landing page, app, game, brand, product demo, or asset manifest asks Codex to create image files, hero images, UI mockups, sprites, diagrams-as-images, thumbnails, or image edits and the operator wants Codex subscription usage instead of API billing.

Create governed Hermes agents as auditable enterprise operating units with manifest, capability stack, memory seed, deployment plan, verification contract, registry entry, and kill switch.

Detect product intent, run SOTA-2026 deepresearch with a capacity-bound parallel or hive budget, define ICPs, and bootstrap or propose/apply taste.md, taste.vision, and ICP artifacts. Use when the user invokes /deepretaste or asks to retaste a product from intent and customer research.

Define the ICP or ICPs for the current product with deepresearch, then draft or apply ICP-driven updates to taste.md and taste.vision. Use when the user invokes /defineicp or asks to tailor the product kernel to ideal customer profiles.

Produce governed recorded product demos with bilingual voiceover, captions, browser evidence, safety gates, and artifact manifests.

Run the full minmaxing workflow with governed hive coordination: metacognitive route, deepresearch, role map, blackboard, packet DAG, aggregation, introspection, verification, and closeout.

Route and calibrate work with model-agnostic metacognitive control. Use when the user asks for thinking-about-thinking, harness effectiveness, task routing, self-correction, confidence calibration, or strategy improvement before execution.

Run a research-backed ICP week-in-the-life product stress test. Use when the user invokes /icpweek or asks to simulate a full week of real-world product usage by an ideal customer and diagnose product, UX, automation, and technical gaps.

Search the local codebase for patterns, symbols, implementations, and references. Use when the user wants to find where code lives or understand how a pattern is implemented without invoking the OpenAI Codex plugin namespace.

minmaxing 5-tier memory system management