devcodex-labs/devcodex

DevCodex – Intent‑driven AI‑Coding Orchestrator

What it is – DevCodex is a Node‑JS command‑line harness that sits on top of existing AI‑coding assistants (GitHub Copilot, Claude Code, Gemini CLI, Grok, Cursor, etc.). It does not replace the model; instead it adds a project‑level engineering layer that:

1. Interprets user intent – turns a natural‑language request into a structured task (type, scope, risk, required skills, verification criteria).
2. Manages project context – reads a generated Profile (project structure, build/test commands, architecture constraints, etc.) and keeps a memory of past tasks, checkpoints, and evidence.
3. Loads the right “Skill” – from a built‑in library of 80+ domain‑specific capabilities (e.g., API design, security review, performance tuning) that are activated only when needed.
4. Runs a controlled workflow – selects one of several engineered work‑flows (dev, fix, analyze, audit, resume, chat) and enforces confirmation points, automated progression, and evidence‑based completion.
5. Supports long‑running tasks – persists task state across sessions, allows safe resumption, and can coordinate across different AI‑coding hosts.
6. Provides evidence‑based hand‑off – before declaring a task finished it reports what was changed, what tests ran, what risks remain, and optionally creates a “self‑evolution” candidate for future reuse.

---

Key Features (as described in the README)

Feature	What it does
Intent‑driven execution	Users only need to state a goal; DevCodex decides the workflow, scope, required skills, and verification steps.
Cross‑host support	Works with six AI‑coding hosts (Codex, Claude Code, Copilot, Gemini CLI, Grok, Cursor) using a common engineering layer.
80+ professional Skills	Pre‑packaged domain knowledge (frontend, backend, security, testing, performance, SRE, etc.) loaded on demand.
Project Profile	Generates a structured “engineer’s handbook” for the repo (stack, build/test commands, architecture boundaries, contracts). The profile is evidence‑based and can be refined over time.
Auto mode	After an explicit @devcodex-auto (or alias) the system can continue without repeated confirmations, while still respecting the original intent and safety checks.
Cross‑session recovery	Task identity, checkpoints, evidence, and remaining work are persisted in .devcodex/ so a new chat can resume exactly where the last left off.
Evidence‑based completion	Reports modifications, test results, build outcomes, remaining risks, and any required next steps before marking a task done.
Controlled self‑evolution	Repeated problems can be turned into reusable Skills after human approval, without automatically overwriting existing rules.
Parallel‑task safety analysis	Before running multiple tasks it checks for file‑level conflicts, shared state, and merge order.
Fine‑grained permission model	Distinguishes read‑only analysis, file modifications, Git commits, pushes, tags, releases, etc., each requiring explicit authorization.
CLI utilities	devcodex init, status, doctor, runtime, governance ledger and other commands for inspection, maintenance, and governance.

---

Who might benefit?

• Individual developers who want AI assistance that respects project conventions and doesn’t lose context between sessions.
• Teams that need reproducible, auditable AI‑generated changes and want to capture recurring engineering knowledge as reusable Skills.
• Organizations looking to integrate multiple AI‑coding tools under a single, policy‑driven workflow without rewriting each tool’s prompts.

---

Maturity & Availability

• Package: Published on npm (devcodex), version badge shown in the README.
• License: AGPL‑3.0.
• Node requirement: >= 18.17.0.
• Documentation: Full docs, getting‑started guide, examples, and API reference hosted at https://devcodex-labs.github.io/devcodex/.
• Status: Actively maintained (the README mentions update procedures, governance ledger, and a roadmap for skill evolution).

---

Quick start (from the README)

# install globally
npm install -g devcodex

# initialise a project (creates .devcodex/, a starter Profile, etc.)
cd /path/to/your/repo
devcodex init

devcodex status   # see current readiness

After init, you can issue a natural‑language task, e.g.:

分析当前项目最值得优先解决的三个问题。
先说明读取范围和证据，只分析，不修改文件。

DevCodex will:

1. Identify the intent, project, and safe read‑only scope.
2. Load relevant Skills.
3. Produce an evidence‑backed analysis without touching code.

---

Where to learn more

• Documentation site: https://devcodex-labs.github.io/devcodex/
• Getting‑started guide: https://devcodex-labs.github.io/devcodex/guide/getting-started
• Examples (cross‑session resume): https://devcodex-labs.github.io/devcodex/examples/resume
• Issue tracker: GitHub Issues page linked from the README.

---

Bottom line

DevCodex is a genuine open‑source project that adds a robust, intent‑driven orchestration layer to existing AI‑coding assistants, turning raw code generation into a disciplined, auditable software‑engineering workflow.