Kimi Code Action

Version updated for https://github.com/xuwenhao/kimi-code-action to version v0.0.2.

  • This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

This GitHub Action automates code review and implementation tasks in pull requests and issues using the kimi-code CLI. It detects which mode to run based on events like comments or mentions of @kimi. Features include automatic mode detection, interactive code assistance, code review, code implementation, progress tracking, commit signing, session resume, and security hardening. The action runs entirely on GitHub runners with minimal external calls.

What’s Changed

First public release of kimi-code-action — agentic GitHub automation driven by the Kimi Code CLI, forked from claude-code-action (MIT, see NOTICE).

What it does

  • @kimi mentions — mention @kimi in issues, PR comments, or reviews; the agent answers questions, implements changes, and pushes branches
  • PR auto-review — tracked review comment with live checkboxes, plus inline comments on exact diff lines (track_progress: true)
  • Custom automation — any prompt on workflow_dispatch / schedule / workflow_run (issue triage, CI failure analysis, …)
  • Commit signing (GitHub API or SSH key), sticky comments, session_id resume
  • Security hardening inherited from upstream — write-permission checks, content sanitization, TOCTOU filtering, config restore from the base branch, restricted git-push, default deny rules for .github/workflows and force-pushes

Usage

- uses: xuwenhao/kimi-code-action@v0
  with:
    kimi_api_key: ${{ secrets.KIMI_API_KEY }}

Requirements

  • KIMI_API_KEY secret — Kimi Code subscribers create keys in the Kimi Code Console (Console › API Keys); Open Platform keys work via kimi_platform: open-cn|open-intl
  • Default model k3 requires a Moderato plan or above; set kimi_model: kimi-for-coding on lower plans

Differences from upstream

  • Runs kimi -p with stream-json parsing; CLI config injected via a generated KIMI_CODE_HOME (permission rules + mcp.json)
  • Not supported: OIDC App-token exchange, WIF/Bedrock/Vertex/Foundry, allowed_non_write_users, structured output (--json-schema), plugins
  • kimi_args maps Claude-style flags to kimi equivalents (--allowedTools, --max-turns, --mcp-config, --append-system-prompt)

Verified end-to-end on real runners: @kimi replies, tracked PR reviews, inline comments. (base) ➜ ~ cat /tmp/kimi-code-action-v0.0.1-release-notes.md First public release of kimi-code-action — agentic GitHub automation driven by the Kimi Code CLI, forked from claude-code-action (MIT, see NOTICE).

What it does

  • @kimi mentions — mention @kimi in issues, PR comments, or reviews; the agent answers questions, implements changes, and pushes branches
  • PR auto-review — tracked review comment with live checkboxes, plus inline comments on exact diff lines (track_progress: true)
  • Custom automation — any prompt on workflow_dispatch / schedule / workflow_run (issue triage, CI failure analysis, …)
  • Commit signing (GitHub API or SSH key), sticky comments, session_id resume
  • Security hardening inherited from upstream — write-permission checks, content sanitization, TOCTOU filtering, config restore from the base branch, restricted git-push, default deny rules for .github/workflows and force-pushes

Usage

- uses: xuwenhao/kimi-code-action@v0
  with:
    kimi_api_key: ${{ secrets.KIMI_API_KEY }}

Requirements

  • KIMI_API_KEY secret — Kimi Code subscribers create keys in the Kimi Code Console (Console › API Keys); Open Platform keys work via kimi_platform: open-cn|open-intl
  • Default model k3 requires a Moderato plan or above; set kimi_model: kimi-for-coding on lower plans

Differences from upstream

  • Runs kimi -p with stream-json parsing; CLI config injected via a generated KIMI_CODE_HOME (permission rules + mcp.json)
  • Not supported: OIDC App-token exchange, WIF/Bedrock/Vertex/Foundry, allowed_non_write_users, structured output (--json-schema), plugins
  • kimi_args maps Claude-style flags to kimi equivalents (--allowedTools, --max-turns, --mcp-config, --append-system-prompt)

Verified end-to-end on real runners: @kimi replies, tracked PR reviews, inline comments.