MyClaude Multi-Agent Workflow
Claude Code plans and verifies while Codex, Gemini or OpenCode backends do the actual coding.
- Source repo
- stellarlinkco/myclaude
- Stars
- ★ 2.8k
- Last updated
- 6mo ago
- License
- AGPL-3.0
- Primary language
- Go
- FA score
- 45/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude CodeClaude API (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Developers already using Claude Code who want code edits and tests run by a different backend such as Codex or Gemini
- Not a fit if
- Solo developers who want a one-command install without configuring backend CLIs
- Users who need a graphical interface rather than terminal slash commands
- Source review
- 45/100 · Major gaps
What does this agent do, and when should you use it?
MyClaude is a multi-agent orchestration workflow built around Claude Code. It splits responsibilities: Claude Code acts as orchestrator for planning, context gathering and verification, while a bundled codeagent-wrapper executes edits and tests through a selectable backend — Codex, Claude, Gemini or OpenCode. An npx installer writes modules, skills and the wrapper into ~/.claude and exposes slash commands such as /do, /omo, /bmad-pilot, /requirements-pilot and /sparv, plus 11 development essentials like /code, /debug and /review. Modules can be toggled in config.json, and --update refreshes installed modules from the latest GitHub release using installed_modules.json. The project is AGPL-3.0 licensed, with a separate commercial licence path for closed-source use.
Running npx github:stellarlinkco/myclaude launches an interactive installer that writes the chosen modules and skills into the target install directory (default ~/.claude): bin/codeagent-wrapper, skills/, agents/, commands/, hooks/, plus auto-generated settings.json and installed_modules.json. Inside Claude Code, the user invokes a workflow command such as /do (five-phase feature development with codeagent orchestration), /omo (multi-agent orchestration with intelligent routing), /bmad-pilot, /requirements-pilot or /sparv (Specify→Plan→Act→Review→Vault), or individual essentials commands like /code and /debug. Claude Code then plans and verifies, delegating concrete edits and test runs to codeagent-wrapper, which drives the selected backend CLI using features such as codex e, --json, -C, resume, --output-format stream-json, -o stream-json, -y and -r. Standalone skills such as browser, codeagent, codex, gemini, product-requirements, prototype-prompt-generator, skill-install and test-cases can be installed separately.
- Developers already using Claude Code who want code edits and tests run by a different backend such as Codex or Gemini
- Teams that want feature development broken into fixed phases (plan, implement, verify) exposed as reusable slash commands
- Engineering organisations applying a BMAD agile workflow with six specialised agents on large projects
- Product managers or solo developers who want to go from requirements to a working prototype quickly
- People who need bulk test-case generation, PRDs or UI prototype prompts without building their own prompt pipeline
- Tech leads standardising daily actions like /code, /review and /refactor across a team
How do you install or deploy this agent?
You need Node.js and a shell. First install at least one backend CLI and confirm it is on PATH:
which codex && codex --version
which claude && claude --version
which gemini && gemini --versionThen run the interactive installer and select the modules you want (for example do and omo) plus codeagent-wrapper:
npx github:stellarlinkco/myclaudeYou can also list installable items, choose an install directory, force overwrite, or update already installed modules:
npx github:stellarlinkco/myclaude --list
npx github:stellarlinkco/myclaude --install-dir ~/.claude --force
npx github:stellarlinkco/myclaude --updateModules are enabled or disabled in config.json:
{
"modules": {
"do": { "enabled": true },
"omo": { "enabled": false },
"bmad": { "enabled": false }
}
}How do you use this agent?
After installation, trigger workflows from Claude Code with slash commands chosen by scenario: /do for feature development, /omo for bug investigation and fixes, /bmad-pilot for large enterprise work, /requirements-pilot for quick prototypes, and /code or /debug for simple tasks. If codeagent-wrapper is reported missing, rerun the installer and select it:
npx github:stellarlinkco/myclaudeIf Codex reports permission denied, set the approval policy in ~/.codex/config.yaml:
approval_policy: "never"If a module fails to load, inspect the install record and force a reinstall:
cat ~/.claude/installed_modules.json
npx github:stellarlinkco/myclaude --forceWhat are this agent's strengths and limitations?
- Clean separation between orchestration (Claude Code) and execution (codeagent-wrapper), with a backend you can switch among Codex, Claude, Gemini and OpenCode
- Several complete workflows ship together — /do, /omo, /bmad-pilot, /sparv, /requirements-pilot — covering different stages from requirements to verification
- The installer supports --list, --update, --install-dir and --force, and modules can be enabled individually so you can install only what you need
- Extra standalone skills such as browser, product-requirements, test-cases and prototype-prompt-generator are available on their own
- Required backend CLI features (e.g. --json, stream-json, resume) are documented per provider so you can check your environment up front
- Depends on Claude Code as the orchestration entry point; it is not a standalone general-purpose agent framework
- Requires pre-installed and authenticated backend CLIs for Codex, Claude, Gemini or OpenCode, with model usage billed to your own accounts
- AGPL-3.0 licensed; closed-source commercial use requires a separate licence from [email protected]
- No OS support matrix is documented, so cross-platform behaviour is not evidenced
- Known quirks (Gemini cannot read files listed in .gitignore, "Unknown event format" notices) must be worked around by the user
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| MyClaude Multi-Agent Workflow This agent | 45 · Major gaps | CLIFree + model costs | ★ 2.8k | 6mo ago | Go | Codex · Claude Code |
| Maestro-Flow | 52 · Major gaps | CLIFree + model costs | ★ 574 | 10d ago | TypeScript | Codex · Claude Code · Claude API |
| OpenTeams | 56 · Major gaps | Desktop appFree + model costs | ★ 581 | 23d ago | TypeScript | Codex · Claude Code |
| Ordewell | 67 · Some gaps | CLIFree + model costs | ★ 209 | today | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
README and the e2e test show subagents run with --dangerously-skip-permissions and Bash(*) full access, and the FAQ advises setting Codex approval_policy to never, which loosens rather than minimizes privilege, so least_privilege is 1; user confirmation is largely absent beyond installer prompts, user_confirmation 1; data flow across multiple backend CLIs and prompt injection is not transparently documented, data_flow_transparency 1; no dedicated sensitive-data section, sensitive_data_handling 1; dependency security only shows pinned CI actions with no audit, dependency_security 1; external effects (writing ~/.claude, overwriting module files) are described but not isolated, external_effects 1; --update overwrites files with no backup or rollback guidance, rollback 1; source attribution has LICENSE and commercial contact but publisher is unverified, source_attribution 1.
Hook unit tests cover activation guards, task state, concurrency and corrupt JSON edge cases, showing self-consistency, self_consistency 2; depends on external codex/claude/gemini/opencode CLIs and Go 1.21 with required features listed but no version pinning or availability checks, dependency_availability 1; tests assert WARN and block reasons, failure_messages 2.
README offers a workflow selection guide and multiple scenario modules, audience_and_scenarios 2; capability boundaries (backend differences, Gemini cannot read .gitignore) are only scattered, capability_boundaries 1; trigger precision relies on slash commands and hook markers without precise trigger conditions, trigger_precision 1; cross-platform CI matrix and install scripts cover multiple environments, environment_fit 2.
README module table and directory structure are clear, information_architecture 2; install commands and config examples are adequate, install_notes 2; version appears as 6.7.0 in package.json but 6.x in the badge, naming_stability 1; FAQ and troubleshooting exist, examples_and_faq 2; known limitations are only scattered FAQ items, known_limitations 1; LICENSE is full AGPL-3.0, license 2; no CHANGELOG, only auto-generated release notes, versioning_changelog 1; maintenance points to GitHub Issues and a commercial email but publisher is unverified, maintenance_responsibility 1.
Outputs are installable modules and commands, output_usability 2; multi-agent orchestration adds some value but overlaps other frameworks, marginal_value 1; the test script sets --max-budget-usd 5 showing cost awareness, but multi-backend cost is unquantified, cost_benefit 1.
README claims partially map to CI and test files, claim_traceability 1; cross-file corroboration is limited to README, package.json, CI and tests, cross_source_corroboration 1; facts and inferences (e.g. 'recommended') are mixed, fact_inference_separation 1.
- The test script uses --dangerously-skip-permissions and Bash(*) full access; assess privilege risk before deployment.
- The FAQ advises setting Codex approval_policy to never, disabling approval safeguards; use with caution.
- --update overwrites module files with no backup or rollback mechanism.
- Publisher identity is unverified by FollowAgents; verify commercial licensing and maintenance commitments independently.
- Depends on multiple external backend CLIs with no pinned version compatibility.