HVE (Hypervelocity Engineering)¶
At a glance¶
| Field | Value |
|---|---|
| Full Name | Hypervelocity Engineering |
| Category | Enterprise AI-Native SDLC |
| Source | microsoft/hve-core |
| Author/Org | Microsoft ISE (Industry Solutions Engineering) |
| License | MIT |
| First Released | 2024 (internal), open-sourced 2025 |
| Current Version | v3.2.2 stable (March 23, 2026); prereleases through v3.3.101 (April 25, 2026) |
| Stars / Popularity | ~1,580 stars · 343 forks |
| Supported Tools | GitHub Copilot (VS Code extension and CLI) |
Compatible coding agents¶
| Agent | Support |
|---|---|
| GitHub Copilot (VS Code) | ✅ Primary — ships as a VS Code extension; repository inventory is larger than the latest verified stable-release inventory |
| GitHub Copilot Coding Agent (github.com) | ⚠️ Partial — instruction files are readable but the RPI workflow requires interactive phase switching |
| Claude Code | ❌ Not supported — tightly coupled to Copilot ecosystem |
| Cursor | ❌ Not supported |
| OpenAI Codex (CLI) | ❌ Not supported |
| Windsurf | ❌ Not supported |
| Gemini CLI | ❌ Not supported |
| Roo Code | ❌ Not supported |
Overview¶
Hypervelocity Engineering (HVE) is Microsoft's enterprise methodology and tooling framework for AI-native application development. The ISE (Industry Solutions Engineering) team created it after delivering a reported 140+ AI solutions in two years. HVE structures the use of AI agents across the software development lifecycle.
HVE targets multidisciplinary enterprise teams of 4–5 people. Its RPI workflow (Research → Plan → Implement → Review) assigns specialized agents to four phases under constraint-based governance. HVE's documentation explains the rationale: "When AI knows it cannot implement, it stops optimizing for 'plausible code' and starts optimizing for 'verified truth.' The constraint changes the goal."
HVE ships as a VS Code extension and a GitHub repository, both built for the GitHub Copilot ecosystem. On October 27, 2026, the main branch contained 71 *.agent.md, 60 *.instructions.md, zero *.prompt.md, and 88 SKILL.md files. The August 8, 2026 counts were 70, 80, 67, and 58, respectively; reusable prompt files have moved to skill packages. These are repository counts, not verified contents of the v3.2.2 stable package.
Pros & cons at a glance¶
| Pros | Cons |
|---|---|
| ✅ Enterprise-grade governance — audit trails, validated artifacts, policy-as-code | ❌ Copilot lock-in — no support for Claude Code, Cursor, or other tools |
| ✅ Research before implementation | ❌ Large agent and instruction inventory can overwhelm small teams |
| ✅ Domain-specific skill collections across the lifecycle | ❌ Manual context clearing between RPI phases adds operational burden |
| ✅ Supply-chain security (SBOM, attestation, CodeQL, OpenSSF) | ❌ Some collections still experimental/preview |
| ✅ Reported enterprise use (AT&T, KT Corporation) | ❌ No autonomous iteration — each phase needs human initiation |
| ✅ Multi-stack support (no-code → pro-code) | ❌ Microsoft ecosystem tilt (Azure, M365, Power Platform) |
HVE adds research, validation gates, audit trails, and policy-as-code for enterprise teams, at the cost of more setup and manual phase transitions.
Core concepts¶
- Velocity-as-Vector — HVE defines velocity as a combination of speed, direction, and quality.
- Constraint-Based Workflows — Repository access controls, environment promotion rules, deployment targets, and policy enforcement define the boundaries within which agents can work.
- Validated Artifacts — Every output, whether from a human or an AI agent, must pass formal validation gates: automated tests, static analysis, security scans, and compliance checks.
- Repeatable Patterns of Integration (RPI) — Standardized integration patterns reduce fragility and make scaling predictable. The RPI acronym also names the core workflow.
- Collection-based architecture — Domain-specific bundles (hve-core, coding-standards, project-planning, security, design-thinking, data-science-engineering, rai, accessibility, and others) that teams adopt selectively. The
.github/skills/tree held 15 top-level collection directories on October 27, 2026, up from the 10 named in earlier documentation; some additions (accessibility, engagement-reporting, hve-slides) are new domains rather than renames.
How it works¶
The RPI workflow divides tasks into four phases. Each uses a dedicated AI agent with specialized instructions, with context cleared between phases.
Scroll horizontally to view the full diagram.
| Phase | Agent | What It Does | Output |
|---|---|---|---|
| Research | Task Researcher | Investigates the codebase, APIs, and documentation. Records evidence and sources, then recommends one approach. | {date}-{topic}-research.md |
| Plan | Task Planner | Creates planning files with checkboxes and line number references. Checks that research exists before proceeding. | Plan and details files |
| Implement | Task Implementor | Executes the plan task by task, verifying each step and recording changes. | Working code + {date}-{topic}-changes.md |
| Review | Task Reviewer | Checks the implementation against the research, plan, and conventions. Runs lint, build, and test suites. | {date}-{topic}-review.md |
Clear context (/clear or new chat) between phases. Each agent has different instructions; prior-phase context can interfere with them.
Beyond the RPI workflow, HVE includes:
| Type | Count on main |
Description |
|---|---|---|
| Agent files | 71 | Specialized AI assistants for research, planning, implementation, and domain tasks |
| Instruction files | 60 | Repository-specific coding guidelines |
| Prompt files | 0 | Reusable task templates — the category that existed on August 8, 2026 (67 files) has been fully migrated |
SKILL.md files |
88 | Self-contained skill packages |
These counts describe the repository on October 27, 2026, not the uninspected v3.2.2 stable VSIX. Contributor docs still support .prompt.md; its absence from the observed tree does not mean the format is retired.
Strengths¶
- Enterprise governance. Constraint-based workflows, validated artifacts, audit trails, and policy-as-code support oversight. Using them does not by itself establish regulatory compliance.
- Research before implementation. A separate Research phase requires investigation and documented findings before planning or code changes.
- Repository coverage. The
mainbranch contains 71 agent files, 60 instruction files, and 88 skills as of October 27, 2026, with no remaining prompt files. Reusable task templates now use Agent Skills packages. - Multi-stack support. Covers no-code (Copilot Studio), low-code (Power Platform), and pro-code (Azure, custom frameworks).
- Supply-chain security. SBOM generation, attestation, OpenSSF Scorecard, and CodeQL analysis address enterprise security requirements.
- Reported enterprise use. HVE grew out of Microsoft's delivery of 140+ AI solutions. AT&T, KT Corporation, and other telcos are reported adopters; these reports are not independent outcome measurements.
Limitations¶
- GitHub Copilot lock-in. Tightly coupled to the Copilot ecosystem. Teams using Claude Code, Cursor, or other AI tools cannot use HVE without significant adaptation.
- Enterprise weight. The large repository inventory can overwhelm solo developers or small teams. The framework is designed for organizations, not individuals.
- Context management burden. The RPI workflow requires manually clearing context between phases and switching agents, adding operational overhead that simpler approaches avoid.
- Experimental collections. The repository marks a dedicated
experimentalcollection, and individual domain collections (security, rai, design-thinking) still carry preview-level content, with stability not guaranteed. - No autonomous iteration. Each phase requires human initiation; HVE does not include a Ralph-style unattended loop.
- Microsoft ecosystem tilt. While not exclusive to Microsoft tools, HVE has a natural bias toward Azure, Microsoft 365, and Power Platform.
Best for¶
- Enterprise teams (4–10+ developers) building production software with governance, compliance, and audit requirements.
- Organizations already invested in GitHub Copilot who want a structured methodology around their existing tooling.
- Multi-stack projects mixing no-code, low-code, and pro-code components that need consistent quality standards.
- Regulated industries (financial services, healthcare, telecom) that require traceability and validated artifacts.
- Teams onboarding new developers who need documented conventions and instructions.
Not ideal for¶
- Solo developers or small teams who find a large enterprise agent and instruction inventory excessive.
- Teams committed to non-Copilot tools (Claude Code, Cursor, Aider) — HVE's tight Copilot coupling makes portability impractical.
- Rapid prototyping or hackathon-style development where governance overhead slows down exploration.
- Projects needing autonomous overnight iteration — HVE's phased approach requires human presence at each phase transition.
- Cost-sensitive teams — the combination of GitHub Copilot subscription and Azure infrastructure may exceed the budget of smaller organizations.
Community & ecosystem¶
The latest verified stable release is v3.2.2 from March 23, 2026. Prereleases have continued past v3.3.101 (April 25, 2026) without a new stable tag. Keep stable releases, prereleases, and unreleased repository activity distinct.
- GitHub: ~1,580 stars and 343 forks on October 27, 2026.
- VS Code Extension: Available on the VS Code Marketplace as "HVE Core" (ise-hve-essentials.hve-core), providing one-click install.
- Microsoft Learn: Featured on the HVE Accelerators Hub.
- Public advocacy: Robin Cole (VP Engineering, Microsoft) has presented at TMForum and on the Telco in 20 podcast (Ep 122). Valentina Alto published an end-to-end Medium walkthrough (April 2026). Paul Yuknewicz called it "rad" for Azure Functions development.
- Enterprise adoption: AT&T (using GitHub Copilot on Azure), KT Corporation (building development capability on Microsoft stack).
- Documentation: The official docs site (microsoft.github.io/hve-core/) covers the framework. Valentina Alto's Medium article and Dave Davis's blog post provide third-party walkthroughs.
- Ecosystem maturity: Stable core with experimental edges. The collection-based architecture allows incremental adoption, but some collections need more time to mature.
Comparison notes¶
vs. Ralph: Opposite ends of the autonomy-governance spectrum. Ralph is a bash loop with minimal structure; HVE has a large repository inventory and constraint-based governance. Ralph is tool-agnostic; HVE is Copilot-locked. Ralph enables AFK runs; HVE requires human presence at phase boundaries.
vs. BMAD: Both are multi-agent methodologies with structured roles. BMAD uses named personas and an agile-oriented module ecosystem. HVE uses specialized agents within a phased RPI workflow. BMAD is more tool-portable; HVE is Copilot-specific and Microsoft-backed.
vs. GSD: GSD Core is a spec-driven methodology with a five-step Discuss → Plan → Execute → Verify → Ship loop. HVE's RPI workflow adds formal research and review phases with enterprise validation gates. GSD is lighter and multi-runtime; HVE is heavier and Copilot-focused.
vs. Superpowers: Both enforce structured development workflows at different scales. HVE uses agents, constraint-based governance, validated artifacts, and audit trails for enterprise environments. Superpowers uses composable behavioral skills for individual developer discipline. HVE is Copilot-only; Superpowers is multi-tool. HVE provides formal governance that Superpowers lacks.
HVE's growth from 11 previously documented skills to 88 SKILL.md files on main, with .prompt.md files dropping to zero, reflects its move toward the skills ecosystem. It does not establish which files shipped in the latest stable VSIX.