Gh Actions Status

An OpenCode plugin that surfaces GitHub Actions workflow run statuses and unresolved PR review comments directly in your editor.

Features

• Instant push detection: Shows a "Waiting for CI..." toast the moment the agent runs git push — before GitHub has even queued the run
• Live status updates: Transitions through Waiting for CI... → 1 running → 1 passing (or 1 failing) as the run progresses
• Post-push agent prompt: When the agent runs git push and CI completes, the plugin automatically prompts the agent with the full run results and any unresolved review comments so it can react autonomously — fixing failures or addressing feedback without being asked
• Unresolved comment count: Toast includes a count of unresolved PR review threads so you always know if there is feedback waiting
• Sidebar panel: Color-coded list of the latest workflow runs for your current branch
• Agent tool: The AI agent can call gh_actions to check CI status and read the full text of unresolved review comments
• Skill file: Bundled SKILL.md is auto-registered via the config hook, with system prompt injection and compaction-safe context so the agent never loses awareness of the tool in long sessions

Requirements

• GitHub CLI (gh) installed and authenticated
• Git available in your PATH

Installation

Add to your opencode.json:

{
  "plugin": ["opencode-gh-actions-status"]
}

Configuration

Pass options as a tuple:

{
  "plugin": [
    ["opencode-gh-actions-status", {
      "branch": "main",
      "limit": 10,
      "workflows": ["ci.yml", "deploy.yml"],
      "pollInterval": 60000,
      "toastInterval": 30000
    }]
  ]
}

Option	Type	Default	Description
branch	string	"current"	Branch to check. "current" uses the active git branch.
limit	number	5	Maximum number of workflow runs to display.
workflows	string[]	all	Filter to specific workflow files.
pollInterval	number	60000	Cache TTL in ms for gh run list results used by the sidebar.
toastInterval	number	30000	How often (ms) the toast poll loop fires while CI runs are active after a push.

How it works

The plugin makes no GitHub API calls at rest. All polling is triggered exclusively by a git push:

1. The tool.execute.after hook detects when the agent runs git push
2. An immediate "Waiting for CI..." toast is shown before GitHub has queued the run
3. A poll loop starts, calling gh run list on toastInterval (default 30s) and updating the toast through each state transition
4. On completion, the final result toast is shown with a 30-minute duration and polling stops
5. The agent session is automatically prompted with the full CI report — including unresolved review comments — so it can autonomously react to failures or reviewer feedback

The toast message format:

1 passing
2 unresolved     ← only shown when there are unresolved PR review threads

Status indicators

Status	Color
success	Green
failure / timed_out	Red
cancelled / skipped	Yellow
in_progress / queued / pending	Blue

Agent tool: gh_actions

When the AI agent calls gh_actions, it receives:

• All workflow runs for the current HEAD commit with status, conclusion, and URL
• Full text of all unresolved PR review threads (file path, line number, author, comment body, and URL)

This allows the agent to proactively check CI results and address code review feedback without you having to ask.

Skill & context persistence

The plugin ships a skills/gh-actions/SKILL.md file that describes when and how to use the gh_actions tool. It is installed automatically via three hooks:

Hook	Purpose
config	Registers the skills/ directory so OpenCode discovers the skill on startup
experimental.chat.system.transform	Injects a short reminder about the tool into the system prompt on every LLM call, surviving context compaction
experimental.session.compacting	Adds context to the compaction summary so the agent retains awareness of the tool after long sessions

Post-push CI prompt

When the agent runs git push (detected via tool.execute.after), the plugin tracks the session ID. Once all workflow runs reach a final state, it calls session.prompt() to inject the full CI report — including unresolved review comments — as a synthetic message. This triggers an agent turn so it can autonomously fix CI failures or address reviewer feedback.

Development

bun install
bun run build
bun run test

For local testing, reference the built plugin in your opencode.json:

{
  "plugin": ["file:///path/to/opencode-gh-actions-status/dist/index.js"]
}