Feature Suggestions for Claude Code GitHub Actions (Inspired by Gemini CLI Actions)

Title: Feature Suggestions for Claude Code GitHub Actions (Inspired by Gemini CLI Actions)

Issue Body:

Summary

With the recent launch of Google's Gemini CLI GitHub Actions, the competitive landscape for AI in CI/CD has evolved. After a comparative analysis, we've identified several powerful features in their offering that, if adopted, could significantly enhance the utility, stability, and enterprise-readiness of the anthropics/claude-code-action.

This issue outlines four key feature suggestions to improve automation, observability, and user control.

---

1. Pre-Packaged, Configurable Workflows

Current State:
The Claude Code Action is a powerful, general-purpose tool that requires users to build workflow logic (triggers, prompts) from scratch for tasks like code reviews or issue management.

Feature Request:
Provide distinct, out-of-the-box, and configurable workflows for common repository tasks, such as automated issue triage.

Competitor Reference:
Google's Gemini CLI Action launched with pre-packaged workflows for "Intelligent Issue Triage" (automated labeling/prioritization) and "Accelerated Pull Request Reviews."

Suggested Functionality:

• Issue Triage Workflow (issue-triage.yml): A template workflow that triggers on issues.opened. It would analyze the issue body to suggest and apply relevant labels (e.g., bug, feature-request, documentation) and optionally assign it to a team or individual based on keywords.

Benefit: This would lower the barrier to entry for repository automation, reduce manual maintenance overhead, and provide immediate value to teams.

---

2. Integrated Observability for the GitHub Action

Current State:
The Claude Code CLI has excellent, well-documented OpenTelemetry support. However, the documentation for the GitHub Action does not specify how to enable or configure this. It's unclear how to pass the necessary OTEL_* environment variables or if a similar settings block exists for the action itself.

Feature Request:
Provide a clear, documented method to stream metrics, logs, and traces from claude-code-action runs to an observability backend.

Competitor Reference:
The run-gemini-cli action includes a settings input with a telemetry block to easily enable this for Google Cloud.

Suggested Functionality:

• An input in action.yml (e.g., enable_telemetry: true).
• A secure method to pass OTEL_* environment variables to the action's runtime (e.g., via a telemetry_config input that accepts secrets).
• Documentation on integrating with at least one popular backend (e.g., OpenTelemetry Collector, Datadog).

Benefit: Enables organizations to effectively monitor cost, token usage, performance, and reliability of Claude within their production CI/CD pipelines.

---

3. Pinning the Underlying CLI Version

Current State:
The action version (e.g., anthropics/claude-code-action@beta) can be pinned, but there is no documented way to pin the version of the underlying @anthropic-ai/claude-code npm package it installs and uses.

Feature Request:
Add an input to the action that allows pinning the specific version of the Claude Code CLI.

Competitor Reference:
The run-gemini-cli action provides a gemini_cli_version input for this exact purpose.

Suggested Functionality:

• A new input in action.yml, such as claude_code_version, that accepts a semantic version (e.g., 1.0.69).

Benefit: This is critical for CI/CD stability. It ensures reproducible workflow runs and prevents unexpected breakages when the underlying CLI is updated with new features or breaking changes.

---

4. High-Level Plan Approval Step

Current State:
Claude Code's permission model operates at the level of individual, sensitive tool calls (e.g., "Allow running npm install?").

Feature Request:
Implement an optional mechanism where the agent can propose a high-level, multi-step plan to solve a request and wait for a single user approval before proceeding with execution.

Competitor Reference:
The Gemini CLI Actions documentation explicitly describes a workflow where the agent can "Post Plan for Approval" and wait for user confirmation.

Suggested Functionality:

1. When given a complex task (e.g., "implement feature X"), Claude first posts its plan as a comment (e.g., "1. Create file A. 2. Add functions B and C. 3. Update tests in file D.").
2. The workflow then pauses and waits for an approval keyword from a user (e.g., @claude proceed).
3. Once approved, Claude executes the entire plan, potentially still asking for fine-grained permissions if not globally approved.

Benefit: This gives users better oversight and control over complex, agentic tasks. It increases trust and prevents Claude from going down an incorrect path, saving both time and API costs.