OMR
GitHub

Models, skills and commands guide

How to read this guide

An agent is a role with its own instructions (plan, research, review, and so on). A model is the AI engine that runs that role. The OmO agents in this inventory carry a default model and a fallback chain when they are declared; if the first is unavailable, the next is tried. When an agent declares no model or chain (for example, the beta-layer Judge), its card says so.

This guide separates two layers. The OmO functionality ported to OpenCode V2 is complete and confirmed. The BETA layer (the Judge agent and the profile skills) is additional and will be polished later.

Default model per agent

Every OmO agent in the inventory has a default model and a fallback chain. This list shows the real model and a short reading of why, by its role and the capabilities, cost and availability it favours. The beta-layer Judge declares no model or chain.

How to change an agent's model

Edit the unified configuration at ~/.omo/omo.jsonc (or the project file .omo/omo.jsonc) and add the agent under the [opencode] block. The model field pins the primary model; the models field sets the ordered chain (the first is primary, the rest are fallbacks).

// ~/.omo/omo.jsonc
{
  "[opencode]": {
    "agents": {
      "librarian": { "model": "google/gemini-3.6-flash" },
      "oracle": { "model": "anthropic/claude-opus-5-5", "reasoning": "max" }
    }
  }
}

An override wins over the default. If you set only model, the provider must be connected (for example via /login or the auth flow); if you want to keep fallbacks, use models with your own order. Hephaestus still requires a compatible GPT provider.

See the agent cards

Providers and primary model

Choosing the primary model and connecting providers is OpenCode's job, not Oh My Rigel's: OpenCode stores the accounts and credentials, and Oh My Rigel only overlays an agent's or category's model and chain.

List the model catalog

OpenCode provides the catalog. This command lists every model available.

opencode models

Choose the primary model

In the TUI, /models chooses the primary model. In opencode.jsonc, the top-level model field sets the project's primary model. The V1 small_model key is not native in V2: it is normalized to the title agent's model (agents.title.model).

/models
{
  "$schema": "https://opencode.ai/config.json",
  "model": "anthropic/claude-sonnet-4-5"
}

Connect a provider

/connect in the TUI lists the available integrations; opencode auth login does the same without the TUI, and the method is chosen in the flow.

/connect
opencode auth login

List connected accounts

Shows the saved accounts and their state; environment connections appear with their type.

opencode auth list

Disconnect a provider

Removes a saved account. Environment connections are not accounts: they are removed by unsetting the variable.

opencode auth logout

Credentials live in OpenCode's database, not in the Oh My Rigel file. To tune an agent's or category's model or chain, edit the [opencode] block in ~/.omo/omo.jsonc (user) or .omo/omo.jsonc (project).

Agents and models

The team that plans, delegates and runs work. Primary agents (they direct the work) come first, then secondary agents (they work for other agents) and finally the Judge from the beta layer.

Primary agents

Sisyphus primary OmO Orchestrator that plans, decides exploration, and delegates across agents and category-with-skills combos.

Function

Orchestrator that plans, decides exploration, and delegates across agents and category-with-skills combos.

When to use it

Lead orchestrator: plans with todos, decides when to explore, and splits work across agents and models. Use it to drive a whole task from start to finish.

Default model
claude-opus-5-5 (max)
Fallback chain
claude-opus-5-5@max → kimi-k3 → gpt-5.6-sol@medium → glm-5.2 → big-pickle
Providers
none declared
Hephaestus primary OmO Autonomous deep worker that executes goals with GPT Codex.

Function

Autonomous deep worker that executes goals with GPT Codex.

When to use it

Autonomous goal-oriented worker that executes with GPT Codex. Use it for deep work when a provider with a supported GPT model is connected; otherwise it is dropped from the roster.

Default model
gpt-6-sol (medium)
Fallback chain
gpt-6-sol@medium → gpt-5.6-sol@medium
Providers
openai, chatgpt-subscription, github-copilot, opencode
Prometheus primary OmO Builds the work plan, with a strict plan format.

Function

Builds the work plan, with a strict plan format.

When to use it

Plan builder: turns a goal into a verified work plan. Use it before execution when the task is broad or ambiguous.

Default model
claude-fable-5-1 (xhigh)
Fallback chain
claude-fable-5-1@xhigh → claude-opus-5-5@max → kimi-k3@max
Providers
none declared
Atlas primary OmO Executor that walks the pending task list and completes it.

Function

Executor that walks the pending task list and completes it.

When to use it

Plan executor: walks the pending task list and completes each item until done. Use it to empty an already written plan.

Default model
claude-sonnet-5
Fallback chain
claude-sonnet-5 → kimi-k3 → gpt-5.6-sol@medium → minimax-m3 → MiniMax-M3 → minimax-m2.7
Providers
none declared

Secondary agents

Oracle secondary OmO Read-only consultant for hard debugging and architecture design.

Function

Read-only consultant for hard debugging and architecture design.

When to use it

Read-only consultant for high-difficulty reasoning: hard debugging and architecture design. Use it when a problem resists or you must choose between alternatives.

Default model
gpt-5.6-sol (xhigh)
Fallback chain
gpt-5.6-sol@xhigh → gpt-5.6-sol@high → gemini-3.1-pro@high → claude-opus-5-5@max → glm-5.2
Providers
none declared
Librarian secondary OmO Researches external code and docs: remote repositories and official documentation.

Function

Researches external code and docs: remote repositories and official documentation.

When to use it

External code researcher: reads remote repositories, retrieves official documentation, and finds usage examples. Use it to understand someone else's library or pattern.

Default model
kimi-for-coding-highspeed (off)
Fallback chain
kimi-for-coding-highspeed@off → gpt-6-luna-fast@low → deepseek-flash@max → qwen3.7-plus → minimax-m2.7 → claude-haiku-4-5
Providers
none declared
Explore secondary OmO Contextual searcher of the codebase: locates where each thing is.

Function

Contextual searcher of the codebase: locates where each thing is.

When to use it

Contextual search inside the codebase: answers where something is and which file implements it. Fire several in parallel to map an unfamiliar module.

Default model
kimi-for-coding-highspeed (off)
Fallback chain
kimi-for-coding-highspeed@off → gpt-6-luna-fast@low → deepseek-flash@max → qwen3.7-plus → minimax-m2.7 → claude-haiku-4-5
Providers
none declared
Multimodal-Looker secondary OmO Reads PDFs, images and diagrams that are not plain text.

Function

Reads PDFs, images and diagrams that are not plain text.

When to use it

Interpreter of non-text material: PDFs, images and diagrams. Use it when the detail lives in a file that is not plain text.

Default model
gpt-5.6-sol (low)
Fallback chain
gpt-5.6-sol@low → kimi-k3 → glm-4.6v → gpt-5-nano
Providers
none declared
Metis secondary OmO Pre-planning analyst: hidden intentions, ambiguities and foreseeable failures.

Function

Pre-planning analyst: hidden intentions, ambiguities and foreseeable failures.

When to use it

Pre-planning consultant: spots hidden intentions, ambiguities and the points where AI usually fails. Use it before planning something complex.

Default model
claude-fable-5-1 (max)
Fallback chain
claude-fable-5-1@max → claude-opus-5-5@max → kimi-k3@max
Providers
none declared
Momus secondary OmO Plan reviewer: judges clarity, verifiability and completeness.

Function

Plan reviewer: judges clarity, verifiability and completeness.

When to use it

Plan reviewer: judges clarity, verifiability and completeness. Use it to challenge a plan before executing it.

Default model
gpt-6-astra (xhigh)
Fallback chain
gpt-6-astra@xhigh → gpt-6-astra@high → gpt-6-astra@high → claude-opus-5-5@max → gemini-3.1-pro@high → glm-5.2
Providers
none declared
Sisyphus-Junior secondary OmO Focused executor with no delegation, same discipline as Sisyphus.

Function

Focused executor with no delegation, same discipline as Sisyphus.

When to use it

Focused executor: the same discipline as Sisyphus but with no delegation. Use it for one bounded unit of work.

Default model
claude-sonnet-5
Fallback chain
claude-sonnet-5 → kimi-k3 → gpt-5.6-sol@medium → minimax-m3 → MiniMax-M3 → minimax-m2.7 → big-pickle
Providers
none declared

Beta layer

Judge primary beta Independent adversarial acceptance auditor (separate beta layer).

Function

Independent adversarial acceptance auditor (separate beta layer).

When to use it

Independent adversarial acceptance auditor: verifies finished work with evidence and returns APPROVED, REJECTED or BLOCKED, without implementing fixes. Use it only when you explicitly ask to judge, audit or test something.

Default model
—
Fallback chain
No chain declared in model-core (beta layer).
Providers
none declared

Skills

Specialized instructions that extend how the agent works on specific tasks. OmO skills first, then the beta layer.

OmO skills

ast-grep OmO Searches and rewrites code by shape (AST) across 25 languages.

Function

Searches and rewrites code by shape (AST) across 25 languages.

Usage

Use it when the target is a syntactic pattern (every call shaped a certain way, a codemod), not literal text.

browser OmO Drives a real browser from the JavaScript evaluation kernel.

Function

Drives a real browser from the JavaScript evaluation kernel.

Usage

Use it for sites you are signed into, forms, JS-rendered pages, screenshots and web QA.

coding-agent-sessions OmO Finds, reads and reconstructs local coding-agent sessions.

Function

Finds, reads and reconstructs local coding-agent sessions.

Usage

Use it to search past conversations, transcripts or runs and recover what an earlier session did.

data-scientist OmO Processes and analyzes data with resident engines (DuckDB, Polars).

Function

Processes and analyzes data with resident engines (DuckDB, Polars).

Usage

Use it for CSV, parquet or JSON, grouping, time series, cleaning or plots.

debugging OmO Runs a hypothesis-driven debugging loop in any language or binary.

Function

Runs a hypothesis-driven debugging loop in any language or binary.

Usage

Use it for hangs, silent failures, wrong responses, memory leaks or reverse engineering.

frontend OmO Builds, styles and polishes web UI and UX.

Function

Builds, styles and polishes web UI and UX.

Usage

Use it for any page, component, layout, animation or visual-quality task.

git-master OmO Handles git work: atomic commits, rebase, squash, blame, bisect and reflog.

Function

Handles git work: atomic commits, rebase, squash, blame, bisect and reflog.

Usage

Use it when a commit or a history investigation is needed; not for ordinary code edits.

init-deep OmO Initializes a hierarchical AGENTS.md knowledge base for a project.

Function

Initializes a hierarchical AGENTS.md knowledge base for a project.

Usage

Use it when a repository needs its structure, commands and conventions documented for agents.

lsp-setup OmO Configures a language server for editor and agent tooling.

Function

Configures a language server for editor and agent tooling.

Usage

Use it when a project needs an LSP or the no-LSP-server notice appears.

programming OmO Applies strict modern practice (typed errors, exhaustive match, tests that can fail) in Python, Rust, TypeScript and Go.

Function

Applies strict modern practice (typed errors, exhaustive match, tests that can fail) in Python, Rust, TypeScript and Go.

Usage

Use it when working on .py, .rs, .ts or .go files.

refactor OmO Guides a refactor, cleanup or restructure with the right decomposition.

Function

Guides a refactor, cleanup or restructure with the right decomposition.

Usage

Use it when asked to refactor, simplify, extract or modernize code.

remove-ai-slops OmO Removes AI-generated code smells behind regression tests.

Function

Removes AI-generated code smells behind regression tests.

Usage

Use it to clean up recent AI-generated changes.

review-work OmO Post-implementation gate review: real manual QA and a single control reviewer.

Function

Post-implementation gate review: real manual QA and a single control reviewer.

Usage

Use it before handing off a PR or when asked to review completed work.

ultimate-browsing OmO Renders, drives and captures web pages, including WAF-blocked hosts.

Function

Renders, drives and captures web pages, including WAF-blocked hosts.

Usage

Use it for JS-rendered sources, persistent logins and screenshots as proof.

ulw-execute OmO Executes a written work plan with state, an evidence ledger and parallel subagents.

Function

Executes a written work plan with state, an evidence ledger and parallel subagents.

Usage

Use it when asked to run a plan saved under .omo/plans.

ulw-plan OmO Planning consultant that writes one complete work plan before coding.

Function

Planning consultant that writes one complete work plan before coding.

Usage

Use it only when the ulw-plan workflow or a pre-implementation plan is requested.

ulw-research OmO Maximum-saturation research with a cooperating team and a cited deliverable.

Function

Maximum-saturation research with a cooperating team and a cited deliverable.

Usage

Use it when research or a deep investigation is requested.

visual-qa OmO Rigorous visual QA across web, terminal and paginated surfaces, with screenshots and a verdict.

Function

Rigorous visual QA across web, terminal and paginated surfaces, with screenshots and a verdict.

Usage

Use it for any UI change or when asked whether something looks right.

Beta layer skills

defuddle beta Extracts clean markdown from web pages, stripping clutter and navigation.

Function

Extracts clean markdown from web pages, stripping clutter and navigation.

Usage

Use it instead of downloading a page when a URL is given to read, summarize or research.

github-public-writing beta Writes or audits public GitHub repository content clearly.

Function

Writes or audits public GitHub repository content clearly.

Usage

Use it for READMEs, public documentation, compatibility tables and repository metadata.

json-canvas beta Creates and edits JSON Canvas (.canvas) files with nodes, edges and groups.

Function

Creates and edits JSON Canvas (.canvas) files with nodes, edges and groups.

Usage

Use it for visual canvases, mind maps and diagrams that connect ideas.

juez-repo beta Audits a GitHub repository the way a recruiter or potential client would see it.

Function

Audits a GitHub repository the way a recruiter or potential client would see it.

Usage

Use it when asked to audit the repository or check whether it is ready to show.

juez-tester beta Diagnoses and fixes a problem, or self-audits finished work before review.

Function

Diagnoses and fixes a problem, or self-audits finished work before review.

Usage

Use it when asked to repair and validate, or to return a self-audit before the audit.

obsidian-bases beta Creates and edits Obsidian Bases (.base) with views, filters and formulas.

Function

Creates and edits Obsidian Bases (.base) with views, filters and formulas.

Usage

Use it to organize notes into tables, cards or lists.

obsidian-cli beta Operates Obsidian vaults from the CLI and supports plugin or theme development.

Function

Operates Obsidian vaults from the CLI and supports plugin or theme development.

Usage

Use it to read, create and search notes, tasks and properties in a vault.

obsidian-markdown beta Creates and edits Markdown with Obsidian syntax.

Function

Creates and edits Markdown with Obsidian syntax.

Usage

Use it for wikilinks, embeds, callouts, properties and tags.

obsidian-second-brain-audit beta Audits a vault so it works as a linked second brain.

Function

Audits a vault so it works as a linked second brain.

Usage

Use it when notes end up disconnected or the vault needs structure.

obsidian-vault-writing-rules beta Enforces professional vault writing: no monoliths, with hierarchy and links.

Function

Enforces professional vault writing: no monoliths, with hierarchy and links.

Usage

Use it in any task that creates or edits documentation inside a vault.

opencode-multi-auth-fix beta Diagnoses account errors from the Codex multi-auth plugin.

Function

Diagnoses account errors from the Codex multi-auth plugin.

Usage

Use it on messages such as no available accounts or an expired token.

prompt-master beta Writes or improves copy-ready OpenCode prompts.

Function

Writes or improves copy-ready OpenCode prompts.

Usage

Use it when asked for a prompt for an OpenCode agent or to improve an existing one.

skill-creator beta Creates new skills, improves existing ones and measures their performance.

Function

Creates new skills, improves existing ones and measures their performance.

Usage

Use it to create a skill from scratch, edit one or optimize its description.

tui-design beta Designs, builds and reviews terminal user interfaces.

Function

Designs, builds and reviews terminal user interfaces.

Usage

Use it to create panels, explorers or interactive terminal views.

Commands

Invocable actions. Each group is a real surface: the Oh My Rigel CLI, the parent product CLI and the in-chat slash commands.

Oh My Rigel CLI (rigel-v2)

rigel-v2 version Prints the Rigel V2 product version.

What it does

Prints the Rigel V2 product version.

What it touches

Read-only over the built version; it changes no state.

rigel-v2 get-local-version Compares the running version with the published one.

What it does

Compares the running version with the published one.

What it touches

Reads the local and remote versions; it changes nothing.

rigel-v2 doctor Runs Rigel V2 health diagnostics across four categories.

What it does

Runs Rigel V2 health diagnostics across four categories.

What it touches

Inspects system, configuration, tools and models; it does not alter state.

rigel-v2 install Installs or refreshes the Rigel V2 surface for a platform.

What it does

Installs or refreshes the Rigel V2 surface for a platform.

What it touches

With --platform opencode it only refreshes the runtime inside the isolated laboratory (opencode-v2-lab); it refuses to move the install outside it and never touches your OpenCode V1. It is not the public route: that is script/agent/setup.sh plus rigel-v2-user-install.mjs install --version 1. The codex and native platforms delegate to their installers.

rigel-v2 uninstall Uninstalls the Rigel V2 surface for a platform.

What it does

Uninstalls the Rigel V2 surface for a platform.

What it touches

With --platform opencode it removes the isolated-laboratory runtime entry and deletes the staged runtime and activation state; it refuses when the config does not register that runtime and never touches your OpenCode V1. The codex and native platforms delegate to their uninstallers.

Aliases

cleanup

rigel-v2 ulw-loop Resolves and prints the Codex ulw-loop component command.

What it does

Resolves and prints the Codex ulw-loop component command.

What it touches

It only resolves and prints the command; it does not run the loop.

rigel-v2 worktree-sweep Sweeps obsolete git worktrees in a repository.

What it does

Sweeps obsolete git worktrees in a repository.

What it touches

By default it removes the merged-and-clean worktrees of the repository; --dry-run only reports. It never sweeps protected prefixes.

rigel-v2 boulder Inspects Rigel V2 native boulder and continuity state.

What it does

Inspects Rigel V2 native boulder and continuity state.

What it touches

Reads and prints the persistent work state; it does not alter it.

rigel-v2 refresh-model-capabilities Resolves the V2 host model catalog into a capabilities snapshot.

What it does

Resolves the V2 host model catalog into a capabilities snapshot.

What it touches

Reads the V2 host catalog and writes the snapshot into the isolated-laboratory Rigel state; it does not touch project code.

rigel-v2 ast-grep Provisions the sg binary for the ast-grep skill.

What it does

Provisions the sg binary for the ast-grep skill.

What it touches

Downloads and places the sg binary; it does not touch project code.

Parent product CLI (omo)

omo install Installs and configures the parent Oh My OpenCode product with a wizard.

What it does

Installs and configures the parent Oh My OpenCode product with a wizard.

What it touches

Configures the parent product; it does not configure Oh My Rigel.

Aliases

setup

omo setup Interactive setup wizard for the parent product.

What it does

Interactive setup wizard for the parent product.

What it touches

Configures the parent Oh My OpenCode product; it is not Oh My Rigel.

omo run Runs OpenCode non-interactively, completing on finished todos and background tasks.

What it does

Runs OpenCode non-interactively, completing on finished todos and background tasks.

What it touches

Starts a session and writes parent-product session state.

omo doctor Checks the health of the parent product installation.

What it does

Checks the health of the parent product installation.

What it touches

Inspects the parent installation and models; it does not change Oh My Rigel.

omo get-local-version Shows the installed parent version and checks for updates.

What it does

Shows the installed parent version and checks for updates.

What it touches

Version read; it modifies no state.

omo config Manages the unified OMO configuration, including migration.

What it does

Manages the unified OMO configuration, including migration.

What it touches

Migrates legacy configuration into ~/.omo/omo.jsonc.

omo cleanup Cleans the managed Codex Light state (cache and marketplace) and repairs project-local Codex artifacts.

What it does

Cleans the managed Codex Light state (cache and marketplace) and repairs project-local Codex artifacts.

What it touches

Only accepts --platform codex; it does not uninstall OmO or Oh My Rigel and does not touch your OpenCode V1.

Aliases

uninstall

omo refresh-model-capabilities Refreshes the models.dev-based model capabilities snapshot.

What it does

Refreshes the models.dev-based model capabilities snapshot.

What it touches

Rewrites the model capabilities cache.

omo version Shows the parent product version information.

What it does

Shows the parent product version information.

What it touches

Read-only.

omo boulder Shows boulder progress, elapsed time and per-task statistics.

What it does

Shows boulder progress, elapsed time and per-task statistics.

What it touches

Reads boulder state.

omo worktree-sweep Reports and, with --apply, removes obsolete git worktrees.

What it does

Reports and, with --apply, removes obsolete git worktrees.

What it touches

By default it only reports (dry-run); --apply removes and prunes. It never removes locked worktrees.

omo ulw-loop Runs the LazyCodex ulw-loop CLI.

What it does

Runs the LazyCodex ulw-loop CLI.

What it touches

Runs the loop and writes its own state.

omo mcp Manages MCP: OAuth login, logout and status.

What it does

Manages MCP: OAuth login, logout and status.

What it touches

Registers or revokes OAuth credentials for MCP servers.

Slash commands (chat)

/goal Sets, shows, pauses, resumes or clears the session's persistent goal.

What it does

Sets, shows, pauses, resumes or clears the session's persistent goal.

What it touches

Writes the current session goal.

/refactor Guided refactor with LSP, ast-grep, architecture analysis and verification.

What it does

Guided refactor with LSP, ast-grep, architecture analysis and verification.

What it touches

May edit project code, always with verification.

/ulw-execute Runs an Atlas work session from a Prometheus plan.

What it does

Runs an Atlas work session from a Prometheus plan.

What it touches

Runs the plan; it may create worktrees and PRs.

/stop-continuation Stops every continuation mechanism in the session.

What it does

Stops every continuation mechanism in the session.

What it touches

Stops continuation loops, todos and boulder; it deletes no finished work.

/remove-ai-slops Removes AI code smells from the branch changes and reviews them.

What it does

Removes AI code smells from the branch changes and reviews them.

What it touches

May edit project files.

/handoff Creates a context summary to continue in a new session.

What it does

Creates a context summary to continue in a new session.

What it touches

Produces a summary; it does not change code.

/hyperplan Adversarial multi-agent planning through team-mode.

What it does

Adversarial multi-agent planning through team-mode.

What it touches

Coordinates several team members to plan; it does not edit code.

/btw Opens an ephemeral side conversation with the agent.

What it does

Opens an ephemeral side conversation with the agent.

What it touches

Creates an isolated side session; it does not touch the main thread.

/side Alias of the side conversation.

What it does

Alias of the side conversation.

What it touches

Same as the side conversation: an isolated session that leaves the main thread alone.