Dev & Engineering opencodecodex-climulti-model-routinglspast-grepmcptmuxagent-orchestration

Oh My OpenAgent

An OpenCode and Codex CLI layer for orchestrated coding workflows and integrated developer tools.

FollowAgents review · FARS-2.1
Not recommended
49/ 100 5-point scale 2.5 / 5
1 2 3 4 5 6
Per-dimension scores and reasoning
1Trust11 / 29 · 1.9/5

Evidence: LSP daemon has authentication and ownership checks, tests cover permission modes (0o700/0o600) and symlink attack protection, indicating handling of sensitive data (e.g., auth tokens). However, overall permission model is not explicit, telemetry (PostHog) is enabled by default, and no user confirmation mechanism is provided. Deductions: telemetry default-on, lack of user confirmation; data flow transparency insufficient; dependency security not fully addressed.

2Reliability6 / 14 · 2.1/5

Evidence: CI workflows (test, typecheck, codex-compatibility) and test files exist, indicating some consistency. However, dependency availability not verified, failure messages not detailed. Deductions: dependency availability not verified; failure messages not detailed.

3Adaptability12 / 18 · 3.3/5

Evidence: README describes multiple use scenarios (Ultimate/Light editions, Team Mode, different models), capability boundaries are stated (e.g., Light edition has only 8 components), trigger method (ultrawork command) is clear, environment fit (OpenCode/Codex) is documented. Deductions: capability boundaries rely on README assertions, not deep code verification.

4Convention10 / 18 · 2.8/5

Evidence: Installation guide, configuration docs, feature docs exist, information architecture clear; install notes detailed; naming has compatibility notes (oh-my-opencode/oh-my-openagent); examples and FAQ present; license is SUL-1.0 (non-standard); version number exists (4.19.4), CI auto-generates changelog; maintenance responsibility clear (author). Deductions: license non-standard, may restrict use; known limitations not explicitly listed.

5Effectiveness7 / 13 · 2.7/5

Evidence: Output usability high (provides many tools and features), marginal value high (solves real problems), but cost-benefit not quantified. Deductions: cost-benefit not quantified, relies on external model subscriptions.

6Verifiability3 / 8 · 1.9/5

Evidence: README includes user testimonials and performance data (e.g., success rate improvement), but no reproducible method provided; cross-source verification insufficient; facts and inferences not clearly separated. Deductions: claims lack traceability, cross-source verification insufficient.

Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision 426657202f7b
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Anonymous telemetry (PostHog) is enabled by default, requiring user opt-out, which may raise privacy concerns.
  • License is SUL-1.0, a non-standard open-source license, which may restrict commercial use.
  • Relies on external model subscriptions (e.g., ChatGPT, Kimi), which may incur significant costs.
Review evidence [1][2][3][4][5][6][7][8]
See the full review method →

What does this agent do, and when should you use it?

Oh My OpenAgent is a software-development agent extension delivered as an Ultimate Edition for OpenCode and a Light Edition for Codex CLI. The OpenCode edition includes named agents such as Sisyphus, Hephaestus, and Prometheus, background agents, Team Mode, 54+ lifecycle hooks, built-in MCPs, and tmux integration. The Codex CLI edition is installed through lazycodex-ai and supplies rules, comment checking, git-bash, LSP, ultrawork, ulw-loop, start-work continuation, telemetry, and plugin-scoped MCPs including grep_app, context7, and codegraph. Both editions expose ultrawork/ulw, LSP support, and rules injection, while agent orchestration and team_* tools are Ultimate-only. It fits developers prepared to install and configure a local CLI extension in a supported agent harness.

Its installers write local plugin and configuration state: bunx oh-my-openagent install for Ultimate and npx lazycodex-ai install for Light. At runtime, it can load applicable context from AGENTS.md, README.md, and .omo/rules/**; Ultimate can also route work by category and coordinate members through tools including team_create, team_send_message, team_task_create, and team_status. It exposes code operations such as lsp_rename, lsp_goto_definition, lsp_find_references, and lsp_diagnostics, alongside AST-Grep, tmux, websearch, context7, and grep_app integrations. ultrawork or ulw starts a shared persistent-work mode, while ulw-loop stores multi-goal state under .omo/ulw-loop/. Ultimate's Hashline editing uses LINE#ID content hashes to validate the lines an edit targets.

  1. An OpenCode team wants a lead agent to delegate frontend, deep-execution, quick-change, and architecture tasks to specialized task categories.
  2. A Codex CLI user wants rules injection, LSP, ultrawork, and MCP components without adopting the full OpenCode edition.
  3. An engineer working in a large repository needs rename, definition lookup, reference search, and diagnostics through LSP tools.
  4. A developer wants /start-work to invoke Prometheus for interview-style planning before implementation begins.
  5. An OpenCode user wants to enable Team Mode and observe a lead plus up to eight parallel members in a tmux layout.
  6. A project maintainer needs to override agent models, permissions, hooks, or Team Mode through .opencode/oh-my-openagent.jsonc.

What are this agent's strengths and limitations?

Pros
  • It has two explicitly scoped delivery modes: a full OpenCode edition and a portable Codex CLI edition with documented feature boundaries.
  • Ultimate includes category-based delegation, background agents, and Team Mode with dedicated team_* coordination tools and tmux visualization.
  • It combines LSP, AST-Grep, MCPs, and hash-validated Hashline edits for code navigation, search, and guarded editing.
  • Rules injection supports AGENTS.md and .omo/rules/**, while configuration can override agents, models, permissions, and hooks.
Limitations
  • Full orchestration, Team Mode, AST-Grep, and tmux integration are limited to OpenCode Ultimate; Codex Light has no team_* tools.
  • Installation and operation require a local shell, filesystem, network access, and a specific host: OpenCode plus Bun for Ultimate, or Codex CLI plus Node/npm for Light.
  • Provider authentication and model selection are part of setup; the documented paths include Anthropic, Gemini, Copilot, Z.ai, and OpenCode Zen, so cost depends on the chosen provider and subscription.
  • Anonymous telemetry is enabled by default; it is described as at most once per machine per UTC day with a hashed installation identifier, but opting out requires configuration or environment variables.

How do you install or deploy this agent?

For OpenCode Ultimate Edition, run bunx oh-my-openagent install, then follow the installer’s model-selection and provider-authentication prompts. For Codex CLI Light Edition, run npx lazycodex-ai install in a Node.js/npm environment; a non-interactive autonomous-permissions setup is npx lazycodex-ai install --no-tui --codex-autonomous. Light does not require Bun and writes its managed plugin, marketplace, and configuration state under ~/.codex/.

How do you use this agent?

After installation, type ultrawork or ulw in the relevant harness to begin the workflow. OpenCode users can enable Team Mode in .opencode/oh-my-openagent.jsonc, setting team_mode.enabled to true and configuring max_parallel_members and tmux_visualization; restart OpenCode to unlock team_* tools. Use /start-work for the Prometheus planning flow, and /init-deep when you want hierarchical AGENTS.md files generated for project context.

How does this agent compare with similar options?

Compared with using Codex CLI alone, the Light Edition adds rules, LSP, git-bash, ultrawork, ulw-loop, and plugin-scoped MCPs, but it does not include Ultimate's agent orchestration or Team Mode. Compared with Claude Code, the project states compatibility with Claude Code hooks, commands, skills, MCPs, and plugins; that compatibility claim does not make Claude Code the runtime host for this product.

FAQ

Do I have to use OpenCode?
No. Ultimate targets OpenCode, while Codex CLI can use Light. The editions have different scopes: Team Mode and agent orchestration are not part of Light.
Does it require a paid model plan?
Setup includes provider authentication and model selection. The project lists several providers and subscription suggestions, but it does not promise a single price or free operation.
Does Light automatically grant full autonomous permissions?
The installer asks whether to configure Codex for autonomous full-permissions mode. It is optional, and --codex-autonomous explicitly enables it in the documented non-interactive command.
Does it send telemetry?
Anonymous active-installation telemetry is enabled by default, described as no more than once per machine per UTC day using a SHA256-hashed installation identifier. Configuration and variables such as OMO_DISABLE_POSTHOG=1 can opt out.
What happens on session or context failures?
The listed features include session recovery for session errors, context-window limits, and API failures; actual recovery still depends on the host, selected model, and provider state.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents