engineering-foundation

Vendor-neutral engineering harness for AI coding agents — one canonical knowledge layer, usable from both Claude Code and OpenAI Codex.

Three portable skills, all following the Agent Skills open standard:

Skill	Purpose
agent-harness	Bootstrap or evolve a repository's AI engineering infrastructure (AGENTS.md/CLAUDE.md, skills, validation, hooks, append-only evolution state)
dev-workflow	Eight gated stages for any nontrivial change: decompose → spec → implement+test → e2e-verify → review → commit → deploy → report
spec-workflow	Numbered tasks under .harness/tasks/ with the full AutoResearch runtime (state: spec/progress/findings/directions/iterations; logs: work/orchestrator/heartbeat), executed autonomously to completion

Install

Claude Code — as a plugin (recommended: versioned, updatable):

/plugin marketplace add YangChengTeam/engineering-foundation
/plugin install harness@engineering-foundation

Skills are then namespaced: /harness:agent-harness, /harness:dev-workflow, /harness:spec-workflow.

Codex — as a plugin:

codex plugin marketplace add YangChengTeam/engineering-foundation

Then open /plugins in Codex CLI and install harness from the Engineering Foundation marketplace (bundled skills load in the next session).

Claude Code + Codex — one-line install (un-namespaced /dev-workflow, $dev-workflow):

curl -fsSL https://raw.githubusercontent.com/YangChengTeam/engineering-foundation/master/scripts/remote_install.sh | bash

From a clone:

git clone https://github.com/YangChengTeam/engineering-foundation.git
cd engineering-foundation && python3 scripts/harness/sync_skills.py

Usage

In any project:

• bootstrap the project's harness: /agent-harness bootstrap (Codex: $agent-harness bootstrap)
• plan a feature: /spec-workflow 新功能… — allocates a numbered spec with persistent state
• implement: /dev-workflow … — or let the agent auto-select from the task
• improve the harness from accumulated evidence: /agent-harness evolve

Repository layout

• SKILL.md — canonical agent-harness skill (root); .harness/skills/ — canonical dev-workflow and spec-workflow
• skills/ — generated plugin copies (do not edit; rebuild with python3 scripts/harness/sync_skills.py)
• scripts/harness/ — deterministic sync / drift-check / validation / spec tooling
• AGENTS.md, CLAUDE.md, REVIEW.md — shared instruction layer with task-to-skill routing

Every runtime-behavior claim in the skills cites a fetched, verified source (see SKILL.md §11).