OpenCode + Oh-My-OpenCode Operator

This skill is an operating guide for using OpenCode as your coding runtime, with Oh-My-OpenCode orchestration through Sisyphus, Prometheus, and Atlas.

What this skill gives you

• One-shot delivery with ulw for focused coding requests.
• Plan-to-execution workflow via @plan and /start-work.
• Repeatable defaults so code quality stays consistent across runs.

Core rules

• Do not edit code directly outside OpenCode unless explicitly asked.
• Prefer Sisyphus-first execution for coding tasks.
• For complex work: plan with Prometheus (@plan) then execute with Atlas (/start-work).

Quick start

Verify Oh-My-OpenCode plugin

cat ~/.config/opencode/opencode.json | grep "oh-my-opencode"

If the command returns output, the plugin is installed.

You can also run the bundled checker:

./scripts/check-omo.sh

Run one-shot coding (Sisyphus + ultrawork)

opencode run --agent sisyphus "ulw implement JWT auth in this service and add tests"

Or use the bundled helper:

./scripts/run-ulw.sh "implement JWT auth in this service and add tests"

Start interactive OpenCode (Sisyphus)

opencode --agent sisyphus

Inside OpenCode:

• Use @plan "..." to invoke Prometheus planning.
• Use /start-work to let Atlas execute the plan.

Failure handling

• If the agent asks clarifying questions mid-implementation, answer them in plan mode (Prometheus) and re-run execution.
• If you need more determinism, re-run with ulw and a smaller, explicit request.

Integration with other skills

• agent-selfie: pair structured coding workflows with generated visual assets.
• gemini-image-gen: use the same workflow discipline for image generation automations.
• agentgram: publish progress updates, findings, and demos produced from your workflow runs.

Changelog

• v0.3.0: Added bidirectional ecosystem links and bundled workflow helper scripts.
• v0.2.0: Reworked positioning, quick start, and metadata for clearer marketplace onboarding.
• v0.1.0: Initial release with Sisyphus/Prometheus/Atlas workflow guidance.