القائمة
Dispatch patterns for OpenCode as the coding engine.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Branch, commit, push, and open PRs with humanized messages, then monitor GitHub Actions CI. Per-project autonomy gating; blocks direct pushes to the default branch and pushes with a dirty tree. Never auto-merges.
Dispatch patterns for Claude Code as the coding engine. Default harness. Implementation dispatches go through dispatch_coder.py (writes the dispatch receipt the commit gate requires).
Testing strategy, TDD enforcement, spec compliance, and regression checks.
Write implementation plans: bite-sized tasks for coding engine dispatch.
Automated fix loop for failed Quality/Reviewer checks. Parses failures, builds escalating prompts, dispatches through the active harness up to 3 times.
Run independent, file-disjoint plan tasks concurrently, each isolated in its own git worktree + branch. Collects per-task results; never auto-merges.
| name | harness-opencode |
| description | Dispatch patterns for OpenCode as the coding engine. |
| version | 1.0.0 |
| metadata | {"hermes":{"tags":["harness","opencode","dispatch"],"related_skills":["implementer","harness-claude-code","harness-antigravity"]}} |
Dispatch patterns for using OpenCode (opencode run) as the coding engine.
opencode run '<prompt>' -m google-vertex/gemini-3.5-flash
The run subcommand executes a message non-interactively and returns the result.
terminal(
command="opencode run '<self-contained task prompt>' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=300
)
terminal(
command="opencode run 'In <file>, update <function> to <change>. Run existing tests to verify.' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=120
)
terminal(
command="opencode run 'Implement <feature> as described: <spec>. Create files in <location>. Write tests in <test-location>. Follow existing patterns in <example-file>.' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=300
)
terminal(
command="opencode run 'Fix bug: <description>. Reproduce with: <repro-steps>. Root cause is likely in <file>. Add a regression test.' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=180
)
terminal(
command="opencode run 'Refactor <component>: <goal>. Preserve all existing behavior. Run tests after each change.' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=300
)
terminal(
command="opencode run '<review prompt>' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=120
)
terminal(
command="opencode run 'You are an independent code reviewer. Review this git diff for security concerns and logic errors.
SECURITY (auto-FAIL): hardcoded secrets, shell injection, SQL injection, path traversal, eval with user input.
LOGIC (auto-FAIL): wrong conditionals, missing error handling for I/O, off-by-one, race conditions.
SUGGESTIONS (non-blocking): missing tests, style, performance, naming.
Run: git diff HEAD~1 HEAD
Report: list security concerns, logic errors, and suggestions.' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=120
)
terminal(
command="opencode run 'Fix ONLY these specific issues. Do NOT refactor or change anything else:
<list of issues>
' --dir <project-dir> --dangerously-skip-permissions -m google-vertex/gemini-3.5-flash",
workdir="<project-dir>",
timeout=120
)
| Flag | Purpose | When to Use |
|---|---|---|
run | One-shot non-interactive mode | Always for coordinator dispatch |
--dir | Set working directory | Always — set project directory |
-m | Override model (e.g., -m google-vertex-anthropic/claude-opus-4-7) | When a specific model is needed for a task |
--variant | Reasoning effort (high, max, minimal) | For complex tasks needing deeper reasoning |
--dangerously-skip-permissions | Auto-approve all tool use | When running unattended via coordinator |
--format json | JSON output for programmatic parsing | When coordinator needs structured output |
-f / --file | Attach file(s) to the prompt | When providing reference files |
--thinking | Show model reasoning blocks | For debugging or complex analysis |
-m--variantCause: Multiple skill folders are named opencode (e.g. under harness/ and autonomous-ai-agents/). The skill_view tool will refuse to guess and throw an ambiguity error.
Solution: Load the skill using its categorized path directly:
skill_view(name="harness/opencode")
google-vertex-anthropic/claude-opus-4-7) configured, which might not be enabled or accessible in your current Vertex AI project region, resulting in a NOT_FOUND Requested entity was not found error on invocation.List the available model list using opencode models to identify accessible Vertex AI models:
opencode models
Override the model on invocation using the -m or --model flag with an active, accessible model (e.g., google-vertex/gemini-2.5-flash or google-vertex/gemini-3.5-flash):
opencode run "..." -m google-vertex/gemini-2.5-flash