By Jiaxin2006
Portable AI engineering harness: skills, agents, and install hooks for Claude Code, Cursor, Trae, and Codex.
Read-only semantic review for the active change set. Use when the user wants a second opinion on correctness, safety, API design, or test gaps before merge.
Classifies issues, CI failures, or bug reports into categories with suggested next steps and which other agent or skill to invoke.
当发现既有 workflow/测试写法不正确或缺失时:自动补齐/新增 skill,并在满足门禁时自动提交并 push。
以考官视角做“任务是否完成”审计:列出验收标准→找不确定点→补测试/证据→修复→重复直到无剩余疑虑。用户要求 100% 确信或准备交付/提 PR 时调用。
Generic experiment workflow guard: sync configured upstream remote and base branch, stop immediately on merge conflicts, prefer containerized verification when POLYGLOT_DOCKER_IMAGE or project docs specify it, and require explicit human approval before delivery/PR actions.
Explains polyglot-agent-harness layout: Claude Code plugin-dir vs .claude copy, Trae skills symlink install, Cursor rules bridge, Codex instructions_file; points to AGENTS.md and docs/PORTABILITY.md.
Structured pull-request preparation: small commits, clear title/body, test evidence, risk notes, and checklist aligned with upstream contribution guidelines when present.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Portable AI engineering harness (skills, agents, commands) aimed at Claude Code (--plugin-dir or copy into .claude/), Trae (.trae/skills), Cursor (.cursor/rules + project skills), and Codex / other tools that read root AGENTS.md.
Before running container workflows, verify the daemon:
python3 scripts/docker-check.py
docker info succeeded (CLI present, daemon reachable).For long Rust/toolchain-heavy workflows, prefer a reusable container or mounted cache volumes instead of repeated docker run --rm. Throwaway containers discard runtime state such as rustup components and cargo downloads, which can make every verification re-download the same toolchain.
Pattern:
docker run -it --name <project>-dev \
-v "$PWD":/workspace \
-v <project>-rustup:/opt/rustup \
-v <project>-cargo-registry:/opt/cargo/registry \
-v <project>-cargo-git:/opt/cargo/git \
-w /workspace \
<image> bash
Resume later with:
docker start -ai <project>-dev
Still run python3 scripts/docker-check.py before Docker workflows. If a project or CI requires a clean container, document that choice and expect repeated dependency/toolchain downloads.
This repository’s hooks/hooks.json uses the nested SessionStart schema compatible with plugins such as starry-harness. The older flat hook list used by tgoskits-plugin supports PreToolUse entries like:
{
"event": "PreToolUse",
"matcher": "Bash",
"command": "python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/docker-check.py\" --pre-tool-use"
}
If your Claude Code install expects that flat format, merge the snippet into your user or project hook settings and point CLAUDE_PLUGIN_ROOT at this repo (or your .claude/ copy). The script’s --pre-tool-use mode only probes Docker when the Bash tool input mentions docker.
| Path | Role |
|---|---|
skills/ | Trae/Cursor-style SKILL.md trees |
agents/ | Claude Code agent prompts |
commands/ | Slash-command markdown |
hooks/ | Plugin lifecycle hooks |
scripts/docker-check.py | Docker daemon health check |
When a new E2E passes, a key bug is fixed, or a PR is being prepared, use the report-generator skill and write in-repo reports under reports/.
See scripts/install-into-project.sh or submodule this repository and symlink skills/* into .trae/skills/ per your team’s convention.
Apache-2.0 — see LICENSE.
npx claudepluginhub jiaxin2006/polyglot-agent-harness --plugin polyglot-agent-harnessFeature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
Production-grade engineering skills for AI coding agents — covering the full software development lifecycle from spec to ship.
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Persistent file-based planning for AI coding agents. Crash-proof markdown plans (task_plan.md, findings.md, progress.md) that survive context loss and /clear, with an opt-in completion gate and multi-agent shared state. Manus-style. Works with Claude Code, Codex CLI, Cursor, Kiro, OpenCode and 60+ agents via the SKILL.md standard. Includes Arabic, German, Spanish, and Chinese (Simplified and Traditional).
Unity Development Toolkit - Expert agents for scripting/refactoring/optimization, script templates, and Agent Skills for Unity C# development
Complete developer toolkit for Claude Code