skills.rest
gabesan21
gabesan21COMMUNITY
Explore all skills by this author →
GitHub

run-opencode

Invokes opencode CLI as a headless executor for delegated coding tasks with NDJSON output.

1|1|Updated Jul 6, 2026
Search Tags:#automation#cli#headless#opencode#ndjson#code delegation#ai coding agent
One-click install
npx skills add https://github.com/gabesan21/project-of-projects --skill run-opencode-gabesan21
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: run-opencode
Source: https://github.com/gabesan21/project-of-projects/tree/main/.agents/skills/run-opencode
Command: npx skills add https://github.com/gabesan21/project-of-projects --skill run-opencode-gabesan21

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Delegating coding work to the opencode CLI from an orchestrator requires knowing the correct headless invocation flags, session handling, yolo-mode permission rules, and output parsing, which vary across versions and are easy to get wrong. ## Core Features & Use Cases - Headless Execution: Runs opencode run with auto-approval, timeouts, and JSON (NDJSON) output for parsing text, tool use, tokens, and cost per event. - Session and Agent Control: Continues, forks, and names sessions, selects built-in or custom agents (build, plan, explore), and picks models across Anthropic, OpenAI, and Google providers. - Use Case: An orchestrator needs tests written for a module: it invokes opencode run --auto --format json in an isolated worktree with a scoped prompt, captures the sessionID from NDJSON events, and extracts token cost from step_finish events. ## Quick Start Ask the agent to run a scoped coding task through opencode in headless mode with auto-approval and JSON output, then parse the NDJSON stream for the final text and cost.

Frequently Asked Questions about run-opencode

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run opencode headlessly from the command line?▼

Use `opencode run --auto --format json "task"` with a timeout wrapper so it processes the prompt, prints events, and exits. Never call bare `opencode` in automation since it opens the interactive TUI, and pass the message as an argument because stdin support is not guaranteed.

How do I parse opencode JSON output for tokens and cost?▼

The `--format json` output is NDJSON with one event per line, including step_start, text, tool_use, and step_finish. Pipe it through jq, selecting `.type=="step_finish"` for tokens and cost, and `.type=="text"` for the final response text.

Does opencode support multiple model providers?▼

Yes, opencode is multi-provider: pass `--model provider/model` such as anthropic/claude-sonnet-4-6, openai/gpt-5, or google/gemini-3-pro. The `--variant` flag controls reasoning effort, and `opencode models` lists available options.

Why does opencode block an action even in yolo mode?▼

An explicit `deny` rule in the repo's opencode.json still wins over the auto-approval flag, and for bash rules the last matching entry wins. Check the repository config for residual deny entries if something is blocked unexpectedly.

What should I do when opencode reports an authentication error?▼

Abort the entire orchestrator task immediately with no retry or fallback, since login is a human precondition handled outside this skill. An immediate failure with no auth signal is an invocation error, not a login problem.

Related Skills

Explore More Software Engineering Skills→
kingpinzskingpinzs

parallel-executor

Orchestrate parallel Claude Code sessions using Git worktrees and detached agents.

Community
Advanced
TyrizxTyrizx

merge

Merge a topic branch into the base branch and resolve conflicts in the main worktree.

Official
Intermediate
angrysky56angrysky56

kanban-dispatcher

Route unclaimed kanban tasks to specialized agent skills via assignee prefix matching.

Community
Intermediate
agent0aiagent0ai
19.1k

orchestrator

Delegates coding tasks to external terminal coding agents via shell execution.

Official
Advanced