Skip to content
/

opsx-feature-dev

opsx-feature-dev · mbertani/opsx-feature-dev · ★ 1 · last commit 2026-04-10

Adds a 7-phase artifact-driven feature development workflow on top of the OpenSpec CLI and Anthropic's official feature-dev plugin.

Best whenEvery phase of feature development should produce an OpenSpec artifact so the work is traceable, resumable, and reviewable.
Skip ifSingle-line bug fixes, Urgent hotfixes
vs seeds
openspecplugin lacks. The explicit…
Primitive shape 5 total
Commands 2 Subagents 3
00

Summary

opsx-feature-dev — Summary

opsx-feature-dev is a Claude Code plugin (also compatible with GitHub Copilot) that adds a structured 7-phase feature development workflow layered on top of the OpenSpec CLI, capturing every decision, design, and task in OpenSpec artifacts for traceability and resumability. The 7 phases are: Discovery (create OpenSpec change), Codebase Exploration, Clarifying Questions, Architecture & Artifacts (proposal.md + design.md + tasks.md), Implementation, Quality Review, and Summary & Archive. Three specialized subagents drive the workflow: code-explorer (traces execution paths and maps architecture layers), code-architect (designs with implementation blueprints), and code-reviewer (confidence-gated review, threshold ≥ 80). The plugin is explicitly a thin layer over Anthropic's official feature-dev plugin and the OpenSpec CLI — it calls openspec instructions and openspec explore/propose/apply/archive commands directly rather than bundling independent skills, and tracks upstream versions via UPSTREAM_VERSION and OPENSPEC_COMPAT files.

Among the seeds, opsx-feature-dev is closest to openspec but goes further by adding a 7-phase structured workflow, three specialized agent personas, and quality review gates that the base openspec plugin lacks. It resembles BMAD-METHOD in using named agent personas (code-explorer, code-architect, code-reviewer) but these are task-scoped subagents within a single workflow, not persistent persona files.

01

Overview

opsx-feature-dev — Overview

Origin

Created by mbertani (GitHub), last pushed 2026-04-10. 1 star. Written in Shell/Markdown as a Claude Code plugin. Explicitly derived from two upstreams: Anthropic's official feature-dev plugin and the OpenSpec CLI.

Philosophy

"Instead of jumping straight into code, this plugin guides you through a 7-phase workflow where every decision, design, and task is captured in OpenSpec artifacts — making the work traceable, resumable, and reviewable."

The key philosophy is artifact-driven documentation: every phase produces or consumes an OpenSpec artifact, ensuring work can be paused mid-workflow and resumed later, reviewed by other developers, and archived with full traceability.

Relationship to Upstreams

The plugin has two explicit upstream dependencies:

  1. OpenSpec CLI — manages the change lifecycle (explore, propose, apply, archive) and generates the structured artifacts
  2. Anthropic's official feature-dev plugin — provides the base command file and the three agent prompt files

The update-from-upstream.sh script and UPSTREAM_VERSION file track which version of the official plugin was last synced. This is an unusual "fork-with-explicit-provenance" pattern.

When to Use / Not Use

Use for:

  • New features touching multiple files
  • Features requiring architectural decisions
  • Complex integrations with existing code
  • Features where requirements are unclear

Do not use for:

  • Single-line bug fixes
  • Trivial, well-defined changes
  • Urgent hotfixes
02

Architecture

opsx-feature-dev — Architecture

Distribution

  • Claude Code plugin: claude plugin marketplace add mbertani/opsx-feature-dev && claude plugin install opsx-feature-dev
  • GitHub Copilot: copilot plugin install mbertani/opsx-feature-dev

Required Runtime

  • Claude Code or GitHub Copilot CLI
  • OpenSpec CLI (brew install openspec)

Repository Structure

opsx-feature-dev/
├── commands/
│   ├── feature-dev.md          # Main workflow command (Claude Code)
│   └── openspec-feature-dev.md # Copilot-adapted version
├── agents/
│   ├── code-explorer.md        # Codebase analysis subagent
│   ├── code-architect.md       # Architecture design subagent
│   └── code-reviewer.md        # Quality review subagent
├── .claude-plugin/
│   └── marketplace.json        # Plugin marketplace manifest
├── .copilot/                   # Copilot-specific configuration
├── OPENSPEC_COMPAT             # OpenSpec CLI compatibility version
├── UPSTREAM_VERSION            # Anthropic official plugin version tracked
├── check-compat.sh             # Compatibility check script
└── update-from-upstream.sh     # Sync script for upstream changes

Dependency Chain

User invokes: /opsx-feature-dev:feature-dev
    └── Calls: openspec instructions (generates artifact structure)
    └── Calls: code-explorer subagent
    └── Calls: openspec explore/propose
    └── Calls: code-architect subagent  
    └── Calls: openspec apply (implementation)
    └── Calls: code-reviewer subagent
    └── Calls: openspec archive (completion)

Target AI Tools

  • Claude Code (primary)
  • GitHub Copilot (secondary, adapted command)
03

Components

opsx-feature-dev — Components

Commands (2)

Name Purpose
feature-dev.md Main 7-phase workflow command for Claude Code
openspec-feature-dev.md Copilot-adapted version (appears as /openspec-feature-dev without namespace)

Subagents / Agents (3)

Name Purpose Tools
code-explorer Deeply analyzes existing codebase features by tracing execution paths, mapping architecture layers, understanding patterns and abstractions, documenting dependencies Glob, Grep, LS, Read, NotebookRead, WebFetch, TodoWrite, WebSearch
code-architect Designs architectures with implementation blueprints, analyzes codebase patterns, proposes approaches Glob, Grep, Read, WebSearch
code-reviewer Reviews for bugs, quality issues, and project convention adherence; applies confidence ≥ 80 threshold before approving Read, Grep, Glob

Scripts (3)

Name Purpose
check-compat.sh Verifies OpenSpec CLI version compatibility with plugin
update-from-upstream.sh Syncs agent files from Anthropic's official feature-dev plugin

Compatibility Tracking Files

File Purpose
OPENSPEC_COMPAT Minimum supported OpenSpec CLI version
UPSTREAM_VERSION Anthropic official plugin version last synced

7-Phase Workflow Phases

Phase Goal OpenSpec Artifact
1. Discovery Understand what to build OpenSpec change created
2. Codebase Exploration Understand existing code Findings summary (via code-explorer)
3. Clarifying Questions Resolve all ambiguities User answers
4. Architecture & Artifacts Design and document proposal.md, design.md, tasks.md
5. Implementation Build the feature Code changes, task checkboxes
6. Quality Review Verify quality Review findings (via code-reviewer, confidence ≥ 80)
7. Summary & Archive Document completion Summary, optional archive
05

Prompts

opsx-feature-dev — Prompts

Excerpt 1 — Main Feature-Dev Command (feature-dev.md)

---
description: Guided feature development with deep codebase understanding, architecture design, 
and OpenSpec artifact-driven documentation. Use when the user wants to build a new feature 
with proper exploration, design, and documentation tracked through OpenSpec changes.
argument-hint: Optional feature description
---

# OpenSpec Feature Development

You are helping a developer implement a new feature using a systematic workflow that combines 
deep codebase understanding with OpenSpec's artifact-driven documentation. Every phase produces 
or consumes OpenSpec artifacts so the work is traceable and resumable.

## Core Principles

- **Ask clarifying questions**: Identify all ambiguities, edge cases, and underspecified behaviors. 
  Ask specific, concrete questions rather than making assumptions. Wait for user answers before 
  proceeding with implementation. Ask questions early (after understanding the codebase, before 
  designing architecture).
- **Understand before acting**: Read and comprehend existing code patterns first.
- **Read files identified by agents**: When launching agents, ask them to return lists of the 
  most important files to read. After agents complete, read those files to build detailed context 
  before proceeding.
- **Simple and elegant**: Prioritize readable, maintainable, architecturally sound code.
- **Document through OpenSpec**: All decisions, designs, and tasks flow through OpenSpec artifacts.
- **Use TodoWrite**: Track all progress throughout the workflow.

Prompting technique: Principle-first behavioral framing. The command begins with six explicit behavioral principles rather than procedural steps, establishing the agent's mindset before describing actions. The "read files identified by agents" principle is a meta-coordination instruction — it tells the main agent to process sub-agent outputs before proceeding.

Excerpt 2 — code-explorer Subagent (agents/code-explorer.md)

---
name: code-explorer
description: Deeply analyzes existing codebase features by tracing execution paths, mapping 
architecture layers, understanding patterns and abstractions, and documenting dependencies 
to inform new development
tools: Glob, Grep, LS, Read, NotebookRead, WebFetch, TodoWrite, WebSearch, KillShell, BashOutput
model: sonnet
color: yellow
---

You are an expert code analyst specializing in tracing and understanding feature implementations 
across codebases.

## Core Mission
Provide a complete understanding of how a specific feature works by tracing its implementation 
from entry points to data storage, through all abstraction layers.

## Analysis Approach

**1. Feature Discovery**
- Find entry points (APIs, UI components, CLI commands)
- Locate core implementation files
- Map feature boundaries and configuration

**2. Code Flow Tracing**
- Follow call chains from entry to output
- Trace data transformations at each step
- Identify all dependencies and integrations

Prompting technique: Role-specialized subagent with explicit toolset. The model: sonnet frontmatter pins the subagent to a specific model tier. The color: yellow is a visual identifier in the agent UI. The subagent has a defined analysis methodology (Feature Discovery → Code Flow Tracing) that functions as a structured investigation protocol.

Excerpt 3 — code-reviewer Confidence Threshold

The code-reviewer agent applies a confidence threshold:

"Reviews for bugs, quality issues, and project convention adherence (confidence >= 80 threshold)"

This threshold-as-quality-gate is a quantitative stop condition: the reviewer does not approve unless it can assign ≥ 80/100 confidence to the implementation's correctness. This prevents over-eager approval and forces the agent to surface specific concerns when confidence is lower.

09

Uniqueness

opsx-feature-dev — Uniqueness

Differs from Seeds

opsx-feature-dev is a thin workflow wrapper over two existing frameworks (OpenSpec and Anthropic's official feature-dev plugin), and its architectural distinctiveness lies in how it handles those dependencies. Unlike openspec (which provides its own commands and skills directly), opsx-feature-dev explicitly calls openspec instructions and openspec apply as shell commands, treating OpenSpec as an external dependency rather than bundling its logic. The UPSTREAM_VERSION + update-from-upstream.sh pattern — tracking and syncing from Anthropic's official plugin — is a "fork-with-provenance" approach not seen in any seed: rather than copying upstream content and diverging silently, this plugin maintains an explicit compatibility contract. The confidence-threshold quality gate (code-reviewer ≥ 80 confidence) is a quantitative quality gate not present in most seeds; BMAD-METHOD uses persona-based review, but opsx-feature-dev's threshold is numerically expressed in the agent description. The combination of code-explorer + code-architect + code-reviewer as three-phase sequential subagents is closest to BMAD's persona approach but scoped to phases rather than roles.

Positioning

opsx-feature-dev fills a gap in the openspec ecosystem: the base openspec plugin provides artifact management but no structured workflow for how to progress from idea to implementation. This plugin adds the 7-phase workflow, subagent delegation, and quality review. It is explicitly intended to "layer on top" of openspec, not replace it.

Observable Failure Modes

  1. Two-dependency fragility: Breaking changes in either OpenSpec CLI or Anthropic's official feature-dev plugin can break the plugin. The check-compat.sh script mitigates but does not eliminate this.
  2. Confidence threshold ambiguity: "confidence >= 80" is described in the code-reviewer's agent description but not enforced mechanically — it relies on the reviewer agent self-assessing its confidence correctly.
  3. No parallelism: Sequential phase execution means each phase must complete before the next starts; there is no parallelization of code-explorer and code-architect.
  4. Minimal adoption: 1 star suggests limited community validation of the workflow design.
04

Workflow

opsx-feature-dev — Workflow

7-Phase Feature Development Workflow

Phase Subagent Involved Artifact Produced Gate
1. Discovery (none) OpenSpec change record None
2. Codebase Exploration code-explorer Execution paths, architecture map, dependencies None
3. Clarifying Questions (main agent) User answers to all ambiguities User must answer
4. Architecture & Artifacts code-architect proposal.md, design.md, tasks.md Architecture choice approval
5. Implementation (main agent via /opsx:apply) Code changes, task checkboxes ticked None
6. Quality Review code-reviewer Review findings (confidence ≥ 80 required) Review threshold gate
7. Summary & Archive (main agent) Summary, optional archive None

Approval Gates

3 gates:

  1. Clarifying Questions (Phase 3): user must answer before proceeding
  2. Architecture Choice (Phase 4): user selects from proposed approaches before implementation
  3. Quality Review Threshold (Phase 6): code-reviewer must reach ≥ 80 confidence

Resumability

Because all decisions are captured in OpenSpec artifacts, the workflow can be paused and resumed:

  • Pause after Phase 4 → resume with /opsx:apply (implementation-only)
  • Artifacts are reviewable independently of the workflow
  • Archive with /opsx:archive to close the change

Platform Compatibility

Feature Claude Code GitHub Copilot
7-phase workflow Yes Yes
code-explorer Yes Yes
code-architect Yes Yes
code-reviewer Yes Yes
OpenSpec integration Yes Yes
Todo tracking TodoWrite SQL todos table
User questions AskUserQuestion ask_user tool
06

Memory Context

opsx-feature-dev — Memory & Context

State Storage

All state is managed by the OpenSpec CLI:

State Type Storage Persistence
Feature change record OpenSpec change (via CLI) Project-level
Codebase findings OpenSpec exploration artifact Project-level
Proposal proposal.md (OpenSpec artifact) Project-level
Technical design design.md (OpenSpec artifact) Project-level
Task breakdown tasks.md (OpenSpec artifact) Project-level
Implementation progress Task checkboxes in tasks.md Project-level
Todo tracking TodoWrite (Claude Code) / SQL todos (Copilot) Session

Cross-Session Resumability

Because all decisions flow through OpenSpec artifacts, the workflow can resume at any phase:

  • After Phase 4: resume with /opsx:apply (skips discovery, exploration, and architecture)
  • Task checkboxes track implementation progress
  • Archive is explicit and optional

Context for Agents

At Phase 2, the code-explorer returns the most important files it found. The main agent is explicitly instructed to read those files before proceeding to Phase 3 (Clarifying Questions). This creates an explicit context-building step between sub-agent output and the next prompt stage.

Memory Type

file-based (OpenSpec artifacts: proposal.md, design.md, tasks.md). No separate memory store.

Spec Storage Pattern

per-feature-folder — each feature gets its own OpenSpec change directory with all artifacts.

07

Orchestration

opsx-feature-dev — Orchestration

Multi-Agent

Yes. Three named subagents: code-explorer, code-architect, code-reviewer. Each runs as a scoped agent for its phase.

Orchestration Pattern

sequential — agents are dispatched one at a time, in phase order. The code-explorer runs first (Phase 2), code-architect runs in Phase 4, code-reviewer runs in Phase 6. There is no parallel fan-out.

Subagent Spawn Mechanism

Subagents are agent definition files (.md with frontmatter). Claude Code spawns them via the agent delegation mechanism. The model: sonnet frontmatter in code-explorer.md pins the subagent model.

Subagent Definition Format

persona-md — each subagent is a markdown file with YAML frontmatter specifying name, description, tools, model, and color.

Isolation Mechanism

none — all agents work in the same project directory. No worktrees, no containers.

Execution Mode

interactive-loop — the workflow waits for user input at three gates (clarifying questions, architecture choice, final review), making it a guided interactive session rather than fully autonomous.

Multi-Model

No — all agents use the same underlying model (Claude, configured to sonnet in subagent frontmatter). The plugin does not route different tasks to different model providers.

Prompt Chaining

Yes — code-explorer's output (list of important files) is explicitly consumed by the main agent before Phase 3. Code-architect's output (proposal.md, design.md) is consumed by the implementation phase via OpenSpec apply instructions. One stage's output IS the next stage's input.

Cross-Tool Portability

Medium — works on Claude Code and GitHub Copilot, not other agent types.

08

Ui Cli Surface

opsx-feature-dev — UI & CLI Surface

Dedicated CLI Binary

No. The plugin uses slash commands invoked from Claude Code or Copilot CLI.

Usage

Claude Code

# Full workflow with description
/opsx-feature-dev:feature-dev Add rate limiting to API endpoints

# Interactive mode
/opsx-feature-dev:feature-dev

GitHub Copilot

/openspec-feature-dev Add rate limiting to API endpoints
/openspec-feature-dev

Note: In Copilot the command appears without namespace prefix.

Local Web Dashboard

No.

Installation Interface

# Claude Code
claude plugin marketplace add mbertani/opsx-feature-dev
claude plugin install opsx-feature-dev

# GitHub Copilot
copilot plugin install mbertani/opsx-feature-dev
# or
gh copilot -- plugin install mbertani/opsx-feature-dev

Observability

  • Workflow progress tracked via TodoWrite
  • OpenSpec artifacts serve as audit trail for every decision
  • UPSTREAM_VERSION and OPENSPEC_COMPAT files track dependency versions

Compatibility Verification

./check-compat.sh

Verifies that the installed OpenSpec CLI version is compatible with the plugin version.

Related frameworks

same archetype · same primary tool · same memory type

Claude-Flow / Ruflo ★ 55k

Eliminates single-agent context limits and sequential bottlenecks by orchestrating fault-tolerant swarms of specialized AI agents…

Hermes Agent (NousResearch) ★ 168k

Self-improving personal AI agent with closed learning loop, 7 terminal backends, and messaging gateway — not tied to any AI…

OpenCode ★ 165k

Terminal-first AI coding agent with multi-model routing, native desktop app, and a typed .opencode/ configuration system for…

OpenHands ★ 75k

Open-source AI software development platform (open-source Devin alternative) with Docker sandbox isolation, 77.6% SWE-bench…

DeerFlow ★ 70k

Long-horizon superagent that researches, codes, and creates by orchestrating parallel sub-agents with isolated contexts in Docker…

oh-my-openagent (omo) ★ 60k

Multi-provider AI agent orchestration for OpenCode: escape vendor lock-in by routing Sisyphus (Claude/Kimi/GLM) and Hephaestus…