Skip to main content

DevHive: Skill-Driven Development (SDD) for OpenCode

DevHive is a fully autonomous AI software development pipeline designed specifically as a set of Skills and a dedicated Agent for OpenCode.

It replaces complex background servers and in-memory states with a transparent, predictable, and local Skill-Driven Development (SDD) architecture. DevHive uses your file system (.devhive/specs/) as its database, generating readable Markdown files at each step of the software lifecycle.

The Pipeline

DevHive is composed of specialized Agent Skills that work sequentially:

  1. devhive-prd: Acts as a Product Manager. Writes or updates a Master docs/PRODUCT_REQUIREMENTS.md and generates a feature-specific .devhive/specs/00-prd.md before any technical work begins.
  2. devhive-explorer: Analyzes the PRD and codebase, then writes .devhive/specs/01-exploration.md
  3. devhive-proposal: Creates technical acceptance criteria in 02-proposal.md
  4. devhive-architect: Designs the system architecture, DB schemas, and UX/UI design system in 03-architecture.md
  5. devhive-taskplanner: Breaks the architecture into actionable items across 8 different domains in 04-tasks.md
  6. devhive-designer: Implements the base UI/UX configuration (CSS vars, Tailwind configs) under ## Design Tasks
  7. devhive-devops: Writes Infrastructure as Code (IaC) (Terraform, CDK, Docker) and checks off ## Infrastructure Tasks
  8. devhive-dba: Creates DB schemas, migrations, and seeders under ## Data Tasks
  9. devhive-backender: Initializes the project, writes APIs and server logic, and checks off ## Backend Tasks
  10. devhive-frontender: Writes UI components and client-side logic, and checks off ## Frontend Tasks
  11. devhive-perf: Writes load testing scripts (k6, artillery) under ## Performance Tasks
  12. devhive-techwriter: Writes API docs (Swagger) and updates README under ## Documentation Tasks
  13. devhive-releaser: Bumps package versions, writes CHANGELOG, and creates local git tags under ## Release Tasks
  14. devhive-sast: Performs a Static Application Security Testing (SAST) scan and writes 05-sast-report.md
  15. devhive-qa: Writes test cases and test plans in 06-qa-plan.md
  16. devhive-auditor: Performs a final check and writes 07-audit.md
  17. devhive-reflection: Extracts persistent memory (semantic, episodic, anti-patterns) to .devhive/memory/ and writes 08-reflection.md

All of this is managed by the devhive-orchestrator skill.

Product Planning Skills

DevHive bundles two skills for turning product requirements into reviewable user stories. They are available after devhive install or devhive update-skill. You can invoke them directly, and @devhive-pm also uses them in batch after generating a Feature PRD.

  • devhive-backlog: Give it PRD text or a readable PRD path to get small, user-centered stories traceable to the source. It calls out important ambiguity rather than inventing requirements. Example: “Use devhive-backlog to break down .devhive/specs/00-prd.md into user stories.”
  • devhive-story: Give it one selected story and, when useful, its source PRD/context. It returns exactly 2–3 Given/When/Then acceptance criteria and a mini plan of at most five bullets, then asks for review. Example: “Use devhive-story to refine US-001 from the approved backlog.”

When used directly, review the backlog before selecting a story for refinement. In the PM workflow, the agent collects all story results into one package and requests a single consolidated review. Approval is required before handoff and never executes a mini plan. These skills are distinct from the PM agent's prioritized docs/PRODUCT/backlog.md artifact (MoSCoW/RICE), which is for product-level prioritization rather than story details.

The Agents

DevHive ships with two OpenCode agents:

@devhive — The Software Engineer

Autonomous software development agent. Hardwired to immediately start the SDD pipeline.

@devhive Build a script that converts CSV to JSON running on AWS Lambda.

@devhive-pm — The Product Manager

Standalone product layer that sits above the SDD pipeline. It defines what to build and why, then hands off to @devhive for how.

@devhive-pm We need a way for customers to track their support tickets.

The PM agent loads the bundled devhive-prd, devhive-explorer, devhive-backlog, and devhive-story skills. It creates the prioritized product backlog before story refinement, then produces at least one story for every in-scope PM-### item (multiple stories when an item represents independent outcomes). Each artifact cross-references PM-### and US-###; items marked Won't have (this iteration) do not require stories.

Artifact Contents
docs/PRODUCT/product-vision.md Problem, audience, value proposition, principles, non-goals
docs/PRODUCT/stakeholders.md Stakeholder map, influence/interest, engagement strategy
docs/PRODUCT/metrics.md North Star metric, supporting KPIs, baselines and targets
docs/PRODUCT/roadmap.md Now / Next / Later horizons with dependencies
docs/PRODUCT/backlog.md Prioritized backlog (MoSCoW + RICE) with effort, value and risk, plus related US-### IDs
docs/PRODUCT/user-stories.md Stories grouped by backlog item and PRD source, each with Given/When/Then criteria, a mini plan of up to five task bullets, and an independent execution status (Pending, In Progress, or Completed); package is Draft until explicitly approved
docs/PRODUCT_REQUIREMENTS.md Master PRD (whole product)
.devhive/specs/00-prd.md Feature PRD for the current cycle

The PM agent never writes code. It asks for one consolidated review of the story package. After the user approves, it marks the package Approved; approval alone does not authorize implementation. On startup, @devhive checks for pending stories and unchecked tasks in .devhive/specs/04-tasks.md. If execution was not explicitly requested, it lists the pending work and asks whether to execute it, select specific items, or defer it. A request to execute all pending work runs eligible stories one at a time in backlog priority order, updating each story's execution status. A request naming one story remains scoped to that story.

@devhive Implement story US-003 from backlog item PM-002 in the approved PM package. Read .devhive/specs/00-prd.md, docs/PRODUCT/backlog.md, and docs/PRODUCT/user-stories.md.

@devhive validates each selected story and maps only its backlog/story IDs and acceptance criteria into technical tasks in .devhive/specs/04-tasks.md; it never defaults to the first story or implements the entire package without explicit authorization. Legacy approved stories with no execution-status field are treated as pending. When invoked directly with no PM PRD/backlog/story package, it keeps the existing self-starting flow: creates its own Feature PRD, then runs the normal SDD phases. PM artifacts are optional inputs; Draft, stale, blocked, or conflicting packages are not silently treated as approved.

Installation

DevHive is available as a Python package via PyPI. You can install it globally to manage your OpenCode skills across all your environments.

pip install devhive

Once installed, you can use the CLI to setup or update your DevHive environment:

# Install DevHive's bundled skills and agents into OpenCode
devhive install

# Update DevHive's bundled skills and agents to the latest versions
devhive update-skill

# Initialize a modern Next.js/React frontend AI guidelines file in your current project
devhive init-frontend

# Initialize a Hexagonal Architecture backend AI guidelines file (Interactive Prompt)
devhive init-backend

DevHive installs only the skills bundled inside this package. It does not download, search for, or install third-party skills from GitHub or other external sources.

The install and update commands also remove the previously installed external shadcn and prd skills, stale DevHive skill directories, remote skill URLs from the OpenCode configuration, and the obsolete DevHive plugin dependency. Other project-local skills under .opencode/skills, .claude/skills, .agents/skills, and skills/ are preserved.

DevHive writes OpenCode 2 configuration under agents.<agent>.permissions and preserves unrelated user configuration. Only the devhive and devhive-pm agents receive the strict DevHive skill allowlist; other agents keep their existing permissions. The bundled devhive-skills-only plugin keeps DevHive's bundled skills, OpenCode built-ins, and project-local skills available while removing external skills from the skill catalog.

Requirements

  • Python 3.7 or newer
  • OpenCode 2 (the bundled devhive-skills-only plugin uses the OpenCode 2 skill-transform API)

Tests

Install the test dependencies and run the test suite:

pip install -e ".[test]"
pytest

Usage (The @devhive Agent)

Instead of using OpenCode as a general assistant, invoke the dedicated @devhive agent. This agent is hardwired to immediately start the SDD pipeline.

@devhive Build a script that converts CSV to JSON running on AWS Lambda.

The agent will automatically load the orchestrator skill and start the 11-phase pipeline!

Continuous Mode

By default, the orchestrator will pause and ask for your approval after writing each specification file (e.g., after writing the architecture design). If you want it to run entirely autonomously without stopping, you can add the "continuous mode" flag to your prompt:

@devhive Build a React landing page. Run in continuous mode with no pauses.

Or permanently enable it for the project by creating a flag file:

mkdir -p .devhive
touch .devhive/continuous

Why SDD (Skill-Driven Development)?

  • Zero Setup: No Python servers, no MCP configuration, no background processes.
  • Local Memory: The entire state of the project is readable by humans in .devhive/specs/. You can edit 02-proposal.md manually, and the Architect will adapt to your changes.
  • Modular: Need a different architectural approach? You can just tweak the prompt inside skills/devhive-architect/SKILL.md.

License

MIT License.

Metadata

Release files for devhive 0.1.46

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for devhive 0.1.46
File Size Uploaded Provenance Actions
devhive-0.1.46.tar.gz 198.7 kB With provenance Details for devhive-0.1.46.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for devhive 0.1.46
File Interpreter ABI Platform Actions
devhive-0.1.46-py3-none-any.whl Python 3 none any Details for devhive-0.1.46-py3-none-any.whl

Total release size: 417.4 kB

Release files / devhive-0.1.46.tar.gz

Download URL devhive-0.1.46.tar.gz
Size 198.7 kB
Tags Source
SHA-256 checksum
How to use checksums
1477e11a07ace4070eebe029c3564f37d5871a1b56cc04611ac55ca4bd29d091
BLAKE2b-256 checksum
How to use checksums
cdbe7053988d8434ab21088fcfe0bd9fb7cca931f7b099cee5f5a507a308b2b3
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 31, 2026.

Transparency log

Release files / devhive-0.1.46-py3-none-any.whl

Download URL devhive-0.1.46-py3-none-any.whl
Size 218.7 kB
Tags Python 3
SHA-256 checksum
How to use checksums
3e6f8361b116a8dec7a2b5ab05eb5c58fe2de4a5d1c8548e04418992be97b75b
BLAKE2b-256 checksum
How to use checksums
a8e9561cc41a0cc77bbdc8a32f60886878ffba5f451eb5493a37598b02a8dfd2
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 31, 2026.

Transparency log

Security for devhive 0.1.46

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for devhive-0.1.46.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 31, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.1.46 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page