codex-eide-rebuild

Rebuild Embedded IDE for VS Code (EIDE) workspaces from Codex or Claude Code and return compact or complete structured JSON build results.

codex-eide-rebuild packages an Agent Skill plus a Windows-first Python runner. The runner reads the current EIDE project model, generates fresh builder.params, invokes EIDE unify_builder through dotnet, and preserves compiler logs, artifacts, memory usage, stack reports, and setup diagnostics for the agent.

Chinese · Agent Skill · Install · Verify · Output · Safety · Plugin · Development

Why

This project is for firmware and embedded teams that already build EIDE projects locally and want agents to run the same rebuild flow without relying on VS Code bridge registration. It keeps the agent-facing protocol simple: one command in, one minimal JSON status out, with the complete JSON saved on disk.

The project is designed for real workspace validation. Agents can run doctor, rebuild all EIDE targets, inspect failed build steps, and quote the exact compiler log path and artifact paths without scraping terminal output by hand.

Highlights

Compatibility

• Windows-first workflow.
• Python 3.11+ with PyYAML.
• Embedded IDE for VS Code (cl.eide) installed in VS Code.
• .NET runtime compatible with the installed EIDE unify_builder.
• EIDE workspaces with .code-workspace and .eide/eide.yml.
• Codex skill installation, a Codex custom agent template, and Claude Code command/subagent templates are included.

Agent Skill and Discovery

This repository includes one Agent Skill:

• Skill name: eide-rebuild
• Skill path: skills/eide-rebuild/SKILL.md
• Primary use case: let an agent rebuild EIDE / Embedded IDE for VS Code firmware workspaces and return structured JSON build results.
• Trigger examples: 帮我编译确认一下, 先 rebuild 看结果, EIDE rebuild C:\work\demo\project.code-workspace, /eide-rebuild C:\work\demo\project.code-workspace.
• Runner entry point: skills/eide-rebuild/scripts/eide_rebuild.py
• Environment check: python skills/eide-rebuild/scripts/eide_rebuild.py doctor
• Codex custom agent template: integrations/codex/agents/eide-rebuild.toml (installed to ~/.codex/agents/eide-rebuild.toml)

Codex subagents are explicit by design. For long or multi-project rebuilds, ask Codex directly: 用 eide-rebuild 子代理 rebuild C:\work\demo\project.code-workspace. If the current Codex surface cannot spawn subagents, the same runner still works directly with minimal JSON stdout.

Tools that index public GitHub repositories for Agent Skills, including SkillsMP-style GitHub indexers, can discover the skill at the path above. This repository uses explicit skill metadata, a stable skill path, and .codex-plugin/plugin.json discovery metadata to make the bundled skill easy to identify.

Project status: community GitHub project with Agent Skill and preparatory Codex plugin metadata. No SkillsMP listing, marketplace listing, or OpenAI official status is claimed.

Plugin Readiness

Codex plugin documentation defines .codex-plugin/plugin.json as the required plugin manifest and keeps skills/ at the plugin root. This repository includes .codex-plugin/plugin.json metadata that points to ./skills/ so plugin-aware tools can identify the bundled skill.

Current supported installation remains the Codex-managed or Claude Code-managed flow in Install. The plugin metadata is discovery and packaging metadata only. Runtime impact: none. Hook installation, permission changes, user configuration edits, background services, and rebuild behavior changes stay out of scope.

Install

Codex

Paste this into Codex:

Fetch and follow instructions from https://raw.githubusercontent.com/gaoguobin/codex-eide-rebuild/main/.codex/INSTALL.md

The install flow clones the repository into ~/.codex/codex-eide-rebuild, installs PyYAML, links the skill namespace into ~/.agents/skills, installs the Codex custom agent template, and runs doctor.

After doctor.ok=true, restart Codex so it rescans installed skills and custom agents.

Claude Code

Paste this into Claude Code:

Fetch and follow instructions from https://raw.githubusercontent.com/gaoguobin/codex-eide-rebuild/main/integrations/claude-code/INSTALL.md

The Claude Code integration installs command and subagent templates. Run /reload-plugins after install.

Update

Codex

Fetch and follow instructions from https://raw.githubusercontent.com/gaoguobin/codex-eide-rebuild/main/.codex/UPDATE.md

Claude Code

Fetch and follow instructions from https://raw.githubusercontent.com/gaoguobin/codex-eide-rebuild/main/integrations/claude-code/UPDATE.md

Restart Codex or run /reload-plugins in Claude Code after updates that change skill files.

Uninstall

Codex

Fetch and follow instructions from https://raw.githubusercontent.com/gaoguobin/codex-eide-rebuild/main/.codex/UNINSTALL.md

Claude Code

Fetch and follow instructions from https://raw.githubusercontent.com/gaoguobin/codex-eide-rebuild/main/integrations/claude-code/UNINSTALL.md

Verify

Run the environment check from the installed skill directory:

python skills/eide-rebuild/scripts/eide_rebuild.py doctor

Run a rebuild against a workspace file or project directory:

python skills/eide-rebuild/scripts/eide_rebuild.py rebuild C:\work\demo\project.code-workspace --stdout minimal

Expected agent behavior:

• Read the minimal JSON status from stdout.
• Treat exitCode=0 as success.
• Treat exitCode=6 as build failure.
• Use other non-zero exit codes for setup, configuration, runtime, or timeout failures.
• Inspect targets[].failureCount, targets[].diagnosticCount, and targets[].artifactCount first.
• Use resultPath for artifact identity, SHA256, complete JSON, deeper compilerLog, steps, or transcript analysis.

Output Protocol

Normal agent runs use minimal stdout:

{
  "schemaVersion": "1",
  "ok": true,
  "exitCode": 0,
  "errorCode": "OK",
  "mode": "rebuild-all",
  "summary": { "discovered": 1, "passed": 1, "failed": 0 },
  "targetNames": ["Debug"],
  "resultPath": "C:/work/demo/build/rebuild_result.json",
  "targets": [
    {
      "name": "Debug",
      "ok": true,
      "exitCode": 0,
      "errorCode": "OK",
      "message": "",
      "failureCount": 0,
      "diagnosticCount": 0,
      "artifactCount": 1
    }
  ]
}

The complete result is always written to resultPath. Use --stdout summary only when stdout needs per-target artifacts, memory, source stats, failures, or diagnostics. Omit --stdout or pass --stdout full only when the full JSON is explicitly needed on stdout.

Common Commands

Agents should use the runner as the source of truth:

python skills/eide-rebuild/scripts/eide_rebuild.py doctor
python skills/eide-rebuild/scripts/eide_rebuild.py rebuild C:\work\demo\project.code-workspace --stdout minimal
python .\scripts\sync_skill_runtime.py --check
python -m unittest discover -s .\runtime\tests -p "test_*.py"

Default paths:

Safety Model

• The runner executes local EIDE rebuild commands in the target workspace.
• The runner writes generated builder.params, compiler logs, artifacts, and build/rebuild_result.json under the project output directories.
• The runner reads project configuration from .code-workspace, .eide/eide.yml, .eide/env.ini, and .eide/files.options.yml.
• The runner may execute pre-build and post-build tasks defined by the EIDE project because those tasks are part of the EIDE build model.
• Out of scope for the runner: VS Code extension installation, VS Code bridge registration, source edits, Codex configuration edits, VS Code settings edits, and hook installation.
• The installer links this repository's skills/ directory into the user's agent skill directory and installs PyYAML for the current user.
• Runtime tests and smoke checks in this repository use sanitized examples and mock data.

Repository Layout

.codex/              Codex install, update, and uninstall guides
.codex-plugin/       Codex plugin discovery manifest
runtime/
  python/            Shared Python runner
  tests/             Unit tests and repository audits
skills/
  eide-rebuild/      Agent Skill and bundled runner copy
integrations/
  codex/            Codex custom agent template
  claude-code/       Claude Code install docs, command, and subagent templates
scripts/
  sync_skill_runtime.py

Development

Sync the runtime into the skill bundle and run tests:

python .\scripts\sync_skill_runtime.py --copy
python .\scripts\sync_skill_runtime.py --check
python -m unittest discover -s .\runtime\tests -p "test_*.py"
python -m compileall runtime\python scripts

Security and Privacy

This repository uses sanitized examples and generic paths only. Runtime tests and smoke checks use mock data. See SECURITY.md for reporting guidance.

Upstream Reference

This project automates rebuild flows for the Embedded IDE for VS Code extension from the upstream EIDE project.

License

MIT - see LICENSE.