Visual Studio Support: Comprehensive Work Item List for Full HVE-Core Parity

[WilliamBerryiii]

HVE-Core currently delivers 55 custom agents, 64 reusable prompts, 104 instruction files, and 12 skill packages to engineering teams through a VS Code Marketplace extension. The extension is purely declarative — zero runtime code, no JavaScript, no activation events. It bundles markdown files and registers them through four VS Code Chat API contribution points (chatAgents, chatPromptFiles, chatInstructions, chatSkills).

Visual Studio and VS Code have converged on the same .github/ file conventions for Copilot customization. The file formats are shared — the gaps are in features, tool naming, and distribution. This discussion catalogs every work item needed to bring the full HVE-Core experience to Visual Studio users.

---

Current State of Visual Studio Support

Artifact Type	VS Code	Visual Studio	VS Version Required
.github/copilot-instructions.md	✅ Full	✅ Full	VS 2022 v17.10+ (opt-in)
.github/instructions/*.instructions.md	✅ Full	✅ Full (including applyTo)	VS 2022 v17.10+ (opt-in)
.github/prompts/*.prompt.md	✅ Full	⚠️ Partial (no frontmatter)	VS 2022 v17.10+
.github/agents/*.agent.md	✅ Full	⚠️ Partial (4 of 13 properties)	VS 2026 v18.4+ only
.mcp.json MCP servers	✅ Full	✅ Full (Tools, Prompts, Resources, Sampling)	VS 2022 v17.14+
Skills (SKILL.md via chatSkills)	✅ Full	❌ Not supported	VS Code extension-only
Planning (multi-step workflows)	Community (via agents)	✅ Native preview (5 tools)	VS 2022 v17.14+ (opt-in)

---

Comprehensive Work Item List

Category 1: Distribution Infrastructure

There is no VS equivalent to the VS Code Marketplace's contributes.chatAgents/chatPromptFiles/chatInstructions/chatSkills contribution points. Artifacts must live in the repo's .github/ directory for VS to discover them.

1.1 — CLI Tool for Artifact Distribution

Build a cross-platform CLI tool that copies versioned HVE-Core artifacts into a project's .github/ directory.

Scope:

• Dual packaging: dotnet tool install hve-core for .NET teams, npx @hve-core/artifacts install for Node.js teams
• Collection selection (e.g., hve-core install --collection coding-standards)
• Maturity filtering (stable vs. pre-release content channels)
• Idempotent install/update (don't clobber user customizations)
• Version metadata for update checks
• Reuse or port the existing Prepare-Extension.ps1 collection resolution and BFS dependency logic
• Exclude agent files from VS 2022-targeted installations (agents require VS 2026)
• Setup guidance for the VS 2022 opt-in checkbox (Tools > Options > GitHub > Copilot > Custom Instructions)

Effort: Medium | Priority: High — this is the minimum viable delivery mechanism

1.2 — MCP Server for Skills and Advanced Agent Behaviors

Build an MCP server exposing HVE-Core domain knowledge as MCP resources and tools.

Scope:

• Convert 12 SKILL.md files to MCP resources (URI-addressable domain knowledge)
• Convert prompt templates to MCP prompt templates with arguments
• Expose agent-like behaviors as MCP tools
• Distribute via npx @hve-core/mcp-server or uvx hve-core-mcp (stdio transport)
• Single .mcp.json entry works in VS, VS Code, Cursor, and JetBrains
• Support SSE/HTTP transports for shared/remote deployments
• Use MCP SDK (TypeScript: @modelcontextprotocol/sdk, Python: mcp)

Effort: High | Priority: High — only mechanism for delivering skills to VS

{
  "servers": {
    "hve-core": {
      "type": "stdio",
      "command": "npx",
      "args": ["@hve-core/mcp-server"]
    }
  }
}

1.3 — Collection Packaging Pipeline

Adapt the existing extension build pipeline for dual output (VSIX + CLI artifact packages).

Scope:

• Extend Prepare-Extension.ps1 to emit a portable artifact bundle alongside the VSIX manifest
• Package artifacts with collection metadata for CLI consumption
• Maintain version parity between VSIX and CLI distributions
• CI/CD pipeline for publishing CLI tool and MCP server packages

Effort: Medium | Priority: Medium

---

Category 2: Agent Frontmatter Gaps (VS Platform Work — Needs VS Team)

HVE-Core agents use 13 YAML frontmatter properties. Visual Studio documents support for only 4: name (optional, falls back to filename), description (required), model (optional), tools (optional, all enabled if omitted). The remaining 9 properties are VS Code-specific with no documented VS equivalent.

Custom agents require VS 2026 v18.4+ — not VS 2022.

2.1 — agents: Frontmatter — Subagent Delegation (P1 Blocker)

HVE-Core's orchestration model depends on agents declaring which other agents they can delegate to:

agents:
  - .github/agents/**/researcher-subagent.agent.md
  - .github/agents/**/plan-validator.agent.md

Without this, the Task Planner cannot invoke the Researcher Subagent, and orchestrator agents cannot delegate work to specialized subagents. This breaks the core RPI (Research-Plan-Implement) workflow.

Impact: Many orchestrator agents | Priority: P1 Blocker

2.2 — handoffs: Frontmatter — Workflow Continuity (P1 Blocker)

Handoffs create guided multi-step workflows by suggesting follow-up actions:

handoffs:
  - label: "Start implementation"
    agent: task-implement
    send: "Implement the plan at .copilot-tracking/plans/"
  - label: "Review changes"
    prompt: /task-review

Supports 5 sub-properties: label, agent, prompt, send, model. Without this, users must manually know which agent or prompt to invoke next.

Impact: 16+ agents | Priority: P1 Blocker

2.3 — disable-model-invocation: Frontmatter (P1 Blocker)

Prevents an agent from responding to user queries directly. Used for utility agents that should only be invoked by parent agents:

disable-model-invocation: true

Without this, 11 internal-only agents appear as user-invocable, creating confusion.

Impact: 11 agents | Priority: P1 Blocker

2.4 — user-invocable: Frontmatter (P1 Blocker)

Controls whether an agent appears in the agent picker. Used to hide subagents and infrastructure agents:

user-invocable: false

Impact: 13 agents | Priority: P1 Blocker

2.5 — argument-hint: Frontmatter (P3 Enhancement)

Provides placeholder text in the chat input when an agent is selected:

argument-hint: "Describe what you want to build or change"

Impact: UX quality | Priority: P3 Enhancement

2.6 — target: Frontmatter (P3 Enhancement)

Undocumented in VS. Low current HVE-Core usage.

Impact: Low | Priority: P3 Enhancement

2.7 — mcp-servers: Frontmatter (P2 Important)

Associates specific MCP servers with an agent:

mcp-servers:
  - hve-core

Impact: Future MCP integration | Priority: P2 Important

2.8 — hooks: Frontmatter (P3 Enhancement)

Preview feature in VS Code. Not documented in VS.

Impact: Low | Priority: P3 Enhancement

---

Category 3: Prompt Frontmatter Gaps (VS Platform Work — Needs VS Team)

VS Code prompts support rich YAML frontmatter. VS documentation shows zero prompt frontmatter properties — prompts are described as plain markdown invoked via #prompt: syntax or the + icon.

3.1 — description: Prompt Frontmatter (P2 Important)

Provides discovery text for the prompt in the prompt picker.

Impact: All 64 prompts | Priority: P2 Important

3.2 — agent: Prompt Frontmatter (P2 Important)

Targets the prompt to a specific agent mode. This is how HVE-Core workflows route users to the right agent configuration:

agent: task-plan

Without this, prompts invoke the default agent instead of the specialized workflow agent. Core workflow entry points depend on agent-targeted prompts.

Impact: Most prompts | Priority: P2 Important

3.3 — argument-hint: Prompt Frontmatter (P3 Enhancement)

Provides placeholder text in the input field when a prompt is selected.

Impact: UX quality | Priority: P3 Enhancement

3.4 — name: Prompt Frontmatter (P3 Enhancement)

Overrides the display name (default is filename-derived).

Impact: 6 prompts | Priority: P3 Enhancement

3.5 — tools: Prompt Frontmatter (P2 Important)

Restricts the tools available when a prompt is active:

tools:
  - read/readFile
  - search/codebase

Impact: 3 prompts | Priority: P2 Important

---

Category 4: Tool Name Incompatibility (VS Platform Work + HVE-Core Work)

VS Code and Visual Studio use completely different tool naming conventions. VS documentation explicitly warns: "Tool names vary across GitHub Copilot platforms."

Known Tool Name Mapping (6 of 33+ tools)

VS Code Tool Name	Visual Studio Tool Name
search/codebase	code_search
read/readFile	readfile
edit/editFiles	editfiles
search/usages	find_references
execute/runInTerminal	runcommandinterminal
web/fetch	getwebpages

VS Code Tools With No Known VS Equivalent

VS Code Tool Name	Purpose	Used By
agent/runSubagent	Delegate to child agent	Orchestrator agents
edit/createFile	Create new files	Implementation agents
edit/createDirectory	Create directories	Implementation agents
search/textSearch	Grep-style text search	Research agents
search/fileSearch	Find files by glob pattern	Research agents
search/listDirectory	List directory contents	Research agents
search/changes	View source control changes	PR and review agents
read/problems	Read Problems panel	Validation agents
read/terminalLastCommand	Get last terminal output	Implementation agents
read/terminalSelection	Get terminal selection	Implementation agents
execute/getTerminalOutput	Get persistent terminal output	Implementation agents
execute/testFailure	Get test failure details	Review agents
vscode/askQuestions	Interactive clarifying questions	Planning agents
todos	Track implementation progress	Implementation agents
selection	Get editor selection	Context-dependent agents

4.1 — Publish Complete VS Tool Inventory (Needs VS Team)

Only 6 VS tool names are publicly documented. HVE-Core needs the full list to build a mapping layer.

Effort: Low (documentation) | Priority: P1 Blocker

4.2 — Cross-Platform Tool Name Mapping or Alias System (Needs VS Team)

Three options:

1. VS recognizes VS Code tool names and maps them internally
2. VS adopts the VS Code naming convention for consistency
3. A cross-platform alias system lets agent authors write tool restrictions once

Without this, agents with tools: restrictions either get no tool access (VS enforces unknown names) or unrestricted access (VS ignores unknown names). Both outcomes are wrong.

Effort: Medium-High | Priority: P1 Blocker

4.3 — HVE-Core Agent Tool Name Compatibility Layer (HVE-Core Work)

Once the full VS tool inventory is published, build a compatibility layer:

• Map VS Code tool names to VS equivalents in agent frontmatter
• Options: parallel agent files per IDE, conditional frontmatter, build-time transformation
• Update the packaging pipeline to emit VS-compatible tools: arrays

Effort: Medium | Priority: P1 — blocked by 4.1

---

Category 5: Skill System Gap (VS Platform Work + HVE-Core Work)

VS Code's chatSkills contribution point has no Visual Studio equivalent. HVE-Core's 12 skills provide deep domain knowledge that agents reference via #file:SKILL.md directives.

Skills Affected

Skill	Domain	Size
python-foundational	Python best practices and idioms	Large
owasp-top-10	Web application security vulnerabilities	Large
owasp-llm	LLM application security vulnerabilities	Large
owasp-agentic	AI agent system security vulnerabilities	Large
pr-reference	Pull request diff analysis and generation	Medium
powerpoint	Slide deck generation with python-pptx	Medium
video-to-gif	FFmpeg video conversion	Small
vscode-playwright	VS Code screenshot capture	Medium
security-reviewer-formats	Security review data contracts	Medium
jira	Jira REST API workflows	Medium
gitlab	GitLab merge request management	Medium
agent-customization	VS Code agent file authoring	Medium

5.1 — MCP Resource Conversion for Skills (HVE-Core Work)

Convert SKILL.md files to MCP resources with URI-addressable content. Part of work item 1.2.

Effort: Medium | Priority: High

5.2 — Native Skill Contribution Mechanism (Needs VS Team)

Introduce a chatSkills-equivalent in Visual Studio, or extend the instructions model to support on-demand (non-auto-attached) domain knowledge delivery.

Effort: High | Priority: P2 Important — MCP resources are a viable workaround

5.3 — #file: Directive Resolution in Prompts and Agents (Needs VS Team)

30+ prompts and 9 agents use #file: directives to compose behavior from shared instruction and skill files:

Read instructions from #file:../../instructions/hve-core/markdown.instructions.md before proceeding.

Ask: Confirm whether Visual Studio resolves #file: directives in prompt and agent body content. If not, prompts that depend on composed instructions will be non-functional.

Effort: Unknown | Priority: P2 Important

---

Category 6: Instructions Improvements (VS Platform Work)

6.1 — Enable Instructions Auto-Attachment by Default (Needs VS Team)

VS 2022 requires an opt-in checkbox: "Enable custom instructions to be loaded from .github/copilot-instructions.md files and added to requests."

This adds onboarding friction — teams must configure each VS installation before instructions work.

Ask: Enable this by default so instructions work without user configuration.

Effort: Low | Priority: P2 Important

---

Category 7: VS Native Capabilities to Leverage

These are VS capabilities that HVE-Core should leverage rather than gaps to fill.

7.1 — VS Native Planning Integration (HVE-Core Work)

VS has native Planning (preview) with 5 dedicated tools: plan, adapt_plan, update_plan_progress, record_observation, finish_plan. HVE-Core currently implements planning through agent orchestration.

Scope:

• Evaluate whether HVE-Core planning agents should consume VS native planning tools
• Determine if the two planning approaches can coexist or if VS-specific agent variants are needed
• Document integration patterns

Effort: Medium | Priority: P3 Enhancement

7.2 — VS Built-in Agents Coordination (HVE-Core Work)

VS has built-in agents (@profiler, @debug, @test, @vs) that overlap with some HVE-Core agent functionality.

Scope:

• Audit overlap between VS built-in agents and HVE-Core agents
• Design coordination patterns (complement vs. replace)
• Consider handoff integration with VS built-in agents

Effort: Low-Medium | Priority: P3 Enhancement

---

Category 8: Testing and Validation (HVE-Core Work)

8.1 — VS Compatibility Test Matrix

Validate each artifact type works correctly in VS:

• Test instructions with applyTo patterns in VS 2022 v17.10+
• Test prompts (plain markdown, no frontmatter) in VS 2022 v17.10+
• Test agents in VS 2026 v18.4+ (when available)
• Test MCP server connectivity via .mcp.json
• Document behavior differences per VS version

Effort: Medium | Priority: High

8.2 — Dual-IDE CI Validation

Extend CI to validate artifacts work in both VS Code and VS contexts:

• Validate tool name arrays contain known VS Code tool names
• Flag agents using unsupported-in-VS frontmatter properties
• Generate compatibility reports per collection

Effort: Medium | Priority: Medium

---

Summary: Work Ownership

Owner	Work Items	Effort Range
HVE-Core Team	1.1, 1.2, 1.3, 4.3, 5.1, 7.1, 7.2, 8.1, 8.2	Medium to High
VS Copilot Chat Team	2.1–2.8, 3.1–3.5, 4.1, 4.2, 5.2, 5.3, 6.1	Medium to High
Both (Coordination)	Tool name mapping (4.1 → 4.3), skill delivery model	Depends on approach

Phased Delivery

Phase	What Ships	VS Coverage	Depends On
Phase 1: Instructions + Prompts	CLI tool distributing instructions and prompts to .github/	VS 2022 v17.10+	1.1, 8.1
Phase 2: Skills via MCP	MCP server exposing skills as resources and tools	VS 2022 v17.14+	1.2, 5.1
Phase 3: Agent Compatibility	Tool name mapping + VS-compatible agent variants	VS 2026 v18.4+	4.1, 4.2, 4.3, 2.1–2.4
Phase 4: Full Parity	All artifact types with full frontmatter support	VS 2026+	All VS team items

---

References

• VS Copilot Chat Context (Microsoft Learn)
• VS Custom/Specialized Agents (Microsoft Learn)
• VS MCP Servers (Microsoft Learn)
• VS Agent Mode (Microsoft Learn)
• HVE-Core Research: VS Code Extension Visual Studio Compatibility

[eugeneboms]

I have two comments:

1. The tools naming mismatch, in my opinion, is important to resolve earlier. It hinders new prompts development by causing confusion.
2. In HVE-DS, there are a few more prompts, some very recent. Long term, should we aim for promoting them into HVE-CORE (provided sufficient use and testing)?