Claude Skills for Copilot
GitHub Copilot agent skills adapted from Anthropic Claude plugins and OpenAI Codex Skills.
Skills
| Skill |
Description |
Source |
/code-review |
Multi-agent code review: bugs, security, conventions-file violations |
Anthropic |
/code-simplifier |
Simplify and refine code for clarity and maintainability |
Anthropic |
/comment-analyzer |
Detect comment rot and misleading documentation |
Anthropic |
/conventions-improver |
Audit and improve AGENTS.md/CLAUDE.md/GEMINI.md quality |
Anthropic |
/feature-dev |
Guided feature development with architecture design and review |
Anthropic |
/frontend-design |
Create production-grade frontend interfaces with high design quality |
Anthropic |
/frontend-layout |
Art-direction-led composition for landing pages and app UI |
OpenAI |
/gh-address-comments |
Fetch PR review comments and address the selected ones via gh CLI |
OpenAI |
/gh-fix-ci |
Debug failing GitHub Actions CI checks and fix after approval |
OpenAI |
/mcp-server-dev |
Guide building MCP servers: deployment model, tool patterns, scaffold |
Anthropic |
/review-toolkit |
Orchestrated end-to-end review (up to 6 specialist sub-skills) |
Anthropic |
/security-threat-model |
AppSec-grade threat model: trust boundaries, abuse paths, mitigations |
OpenAI |
/silent-failure-hunter |
Find silent failures, swallowed errors, and unjustified fallbacks |
Anthropic |
/test-analyzer |
Behavioral test coverage review — gaps, not just line coverage |
Anthropic |
/type-design-analyzer |
Type design quality: invariants, encapsulation, enforcement |
Anthropic |
Usage
Slash command — type / to see available skills, then select or type the skill name (e.g., /code-review).
Auto-invocation — in agent mode, Copilot automatically uses skills relevant to your task.
By instruction — explicitly tell the agent in natural language (e.g., "Please review code changes by code review skill").
Adaptation Details
Skills are adapted from two open-source repositories, both rewritten for VS Code Copilot agent skills format.
From Anthropic Claude Plugins
Source: anthropics/claude-plugins-official (Apache 2.0). The original plugins target Claude Code (CLI).
Common changes
Every skill from this source went through the following transformations:
| Change |
Original (Claude Code) |
Adapted (VS Code Copilot) |
| File format |
agent/command YAML frontmatter with model:, allowed-tools: |
SKILL.md with name, description, argument-hint, user-invocable |
| Tooling |
Bash(gh pr view …), TodoWrite, $ARGUMENTS |
Generic VS Code Copilot tool references |
| Description field |
Long XML <example> chat transcripts embedded in frontmatter |
Single-sentence description |
| Convention file |
CLAUDE.md |
AGENTS.md or similar (CLAUDE.md, GEMINI.md) with priority AGENTS.md -> CLAUDE.md -> GEMINI.md |
| Model names |
claude-haiku-4-5, claude-sonnet-4-5, claude-opus-4-5 |
Generic sub-agent references |
/code-review
Source: plugins/code-review/commands/code-review.md
The original is a PR-focused command that reads the diff via gh pr diff and spawns multiple Haiku sub-agents to review different aspects. It is tightly coupled to GitHub workflow — it requires an open PR and the gh CLI.
Changes:
- Removed all
gh CLI usage; the skill now works on any file, directory, diff, or codebase passed as an argument
- Replaced named Anthropic model invocations with generic sub-agent calls
- Replaced single-file CLAUDE.md convention lookup with AGENTS.md or similar conventions files (
CLAUDE.md, GEMINI.md) in priority order: AGENTS.md -> CLAUDE.md -> GEMINI.md
/code-simplifier
Source: plugins/code-simplifier/agents/code-simplifier.md
The original agent is scoped to "recently modified files" and references Anthropic-internal project conventions (ES modules, arrow functions, specific test patterns). The model: claude-opus-4-5 instruction makes it a high-quality but heavyweight agent.
Changes:
- Removed hardcoded project-specific coding standards from the prompt body
- Broadened scope from "recently modified code" to any file, diff, or codebase segment specified by the user
- Converted from agent (with
model:) to a skill
Source: plugins/pr-review-toolkit/agents/comment-analyzer.md
The original agent embeds lengthy XML <example> transcripts inside the description frontmatter field to illustrate expected output. The body mixes free-form instruction with example scenarios.
Changes:
- Stripped all XML examples from the frontmatter description
- Restructured the body as a numbered analysis framework (comment types → accuracy check → completeness check → maintainability check → scoring)
- Added
argument-hint for user-facing guidance
/conventions-improver
Source: plugins/claude-md-management/skills/claude-md-improver/SKILL.md
The original skill audits and improves CLAUDE.md files in Claude Code projects. It references Claude Code–specific file types (.claude.local.md, ~/.claude/CLAUDE.md), the # key shortcut for incorporating learnings, and uses tool declarations (Read, Glob, Grep, Bash, Edit). Quality criteria and templates are stored in separate references/ files.
Changes:
- Renamed from
claude-md-improver to conventions-improver to reflect the broader scope
- Generalised from
CLAUDE.md-only to support AGENTS.md, CLAUDE.md, and GEMINI.md with priority order AGENTS.md -> CLAUDE.md -> GEMINI.md
- Removed Claude Code–specific file types (
.claude.local.md, global ~/.claude/CLAUDE.md, the # shortcut tip)
- Removed tool declarations from YAML frontmatter
- Embedded quality-criteria and templates content inline (single self-contained
SKILL.md rather than split across references/ files)
/feature-dev
Source: plugins/feature-dev/commands/feature-dev.md
The original command uses the Claude Code–specific TodoWrite built-in tool to manage a persistent task list, and receives the feature description via the $ARGUMENTS shell variable.
Changes:
- Replaced
TodoWrite invocations with a generic "maintain a todo list" instruction
- Replaced
$ARGUMENTS with a natural-language argument hint in the frontmatter
- Converted from command to skill format
/frontend-design
Source: plugins/frontend-design/skills/frontend-design/SKILL.md
This is the closest to a direct port — the source is already a SKILL.md intended for the Claude Code skill system, which shares a similar format to VS Code Copilot skills.
Changes:
- Removed the
license: Complete terms in LICENSE.txt frontmatter field (not used by VS Code Copilot)
- Added
argument-hint and user-invocable: true frontmatter fields
- Minor formatting normalisation (bullet indentation)
Sources:
The original is a PR review orchestrator: it reads a PR with gh pr view, then spawns the 6 specialist agents by filename. Each agent file is standalone and referenced by path.
Changes:
- Completely rewritten: dispatches the other 5 Copilot skills (not agent files) by skill name
- Removed all PR/GitHub context; works on any codebase, directory, or diff
- Replaced
gh pr view with instructions to infer scope from argument or current workspace
- Added a summary table format for the final aggregated report
/silent-failure-hunter
Source: plugins/pr-review-toolkit/agents/silent-failure-hunter.md
The original agent has an extensive description field with XML <example> blocks showing sample review dialogues. The body defines detection patterns for swallowed errors and missing fallbacks.
Changes:
- Stripped XML examples from the frontmatter description
- Restructured the body as a numbered review process with explicit confidence threshold (report only if confidence ≥ 80)
- Added
argument-hint for user-facing guidance
/test-analyzer
Source: plugins/pr-review-toolkit/agents/pr-test-analyzer.md
The original is named pr-test-analyzer and is framed around reviewing tests in the context of a pull request diff. It focuses on whether new tests were added for changed code.
Changes:
- Renamed from
pr-test-analyzer to test-analyzer to reflect the broader, non-PR scope
- Removed all PR framing; the skill now reviews any test file, directory, or codebase
- Added DAMP (Descriptive And Meaningful Phrases) principles as an explicit evaluation criterion
- Added
argument-hint for user-facing guidance
/type-design-analyzer
Source: plugins/pr-review-toolkit/agents/type-design-analyzer.md
The original agent has an informal instruction style and uses XML <example> transcripts in its description to illustrate expected output quality.
Changes:
- Replaced verbose description with a single-sentence summary
- Reformatted the analysis framework as numbered steps with a 4-dimension rating rubric (invariants, encapsulation, enforcement, expressiveness; each scored 1–10)
- Added
argument-hint for user-facing guidance
/mcp-server-dev
Source: plugins/mcp-server-dev/skills/build-mcp-server/SKILL.md
The original skill is structured around Claude Code's tool ecosystem and references six companion references/ files (remote-http-scaffold, deploy-cloudflare-workers, tool-design, auth, elicitation, server-capabilities). It also hands off to two peer skills (build-mcp-app, build-mcpb) that exist in the same plugin but are not yet adapted.
Changes:
- Renamed from
build-mcp-server to mcp-server-dev to match the parent plugin name and our naming convention
- Removed
version: frontmatter field (not used by VS Code Copilot)
- Added
argument-hint and user-invocable: true frontmatter fields
- Removed all
references/xxx.md internal file pointers; Phase 5 now includes inline scaffolding guidance for each deployment path, and replaced the References section with links to public MCP SDK repositories and the MCP specification
- Removed hand-off instructions for
build-mcp-app and build-mcpb (those sub-skills don't exist in this repo); replaced with inline guidance to continue implementing the chosen approach
From OpenAI Codex Skills
Source: openai/skills (Apache 2.0 per-skill). The original skills target the OpenAI Codex agent.
Common changes
| Change |
Original (OpenAI Codex) |
Adapted (VS Code Copilot) |
| File format |
Separate agents/openai.yaml for UI config (display name, icons, prompt) |
Removed; argument-hint and user-invocable: true added to SKILL.md |
| Bundled scripts |
Python scripts/ helpers called by the workflow (gh-fix-ci, gh-address-comments) |
Replaced with inline gh CLI commands |
| Reference files |
Companion references/ files carrying output contracts and guidance (security-threat-model) |
Inlined into SKILL.md |
| Platform-specific |
create-plan skill, sandbox_permissions=require_escalated auth commands |
Removed |
Source: skills/.curated/gh-address-comments/SKILL.md
The original skill uses a bundled Python script (scripts/fetch_comments.py) to fetch PR comments via the GitHub GraphQL API. It also refers to OpenAI Codex–specific auth sandbox commands (sandbox_permissions=require_escalated).
Changes:
- Replaced the Python script call with an inline
gh api graphql command block that fetches conversation comments, reviews, and review threads directly
- Added a REST fallback (
gh pr view --json reviews,comments) for environments where GraphQL is unavailable
- Removed Codex sandbox-specific auth instructions
- Added
argument-hint and user-invocable: true frontmatter fields
/gh-fix-ci
Source: skills/.curated/gh-fix-ci/SKILL.md
The original skill orchestrates a Python script (scripts/inspect_pr_checks.py, 509 lines) to fetch failing check logs. It also references the Codex-specific create-plan skill for the approval step.
Changes:
- Replaced the Python script call with the manual
gh CLI commands already present in the original as a fallback (the script is a convenience wrapper around the same gh run view and gh api calls)
- Replaced
create-plan skill invocation with an inline plan-and-approval step
- Removed Codex sandbox-specific auth instructions
- Added
argument-hint and user-invocable: true frontmatter fields
/security-threat-model
Source: skills/.curated/security-threat-model/SKILL.md
The original skill references two companion files (references/prompt-template.md and references/security-controls-and-assets.md) that carry the output contract and quality rules.
Changes:
- Inlined the output format contract and evidence-grounding rules from
prompt-template.md directly into SKILL.md (VS Code skills support only a single SKILL.md file)
- Inlined the Mermaid diagram rules and required output section list
- Removed references to the separate
references/ directory
- Added
argument-hint and user-invocable: true frontmatter fields
/frontend-layout
Source: skills/.curated/frontend-skill/SKILL.md
The original skill is already in SKILL.md format and focuses on art direction, composition restraint, and image-led hierarchy — complementing the existing /frontend-design skill which emphasises bold aesthetic direction.
Changes:
- Renamed from
frontend-skill to frontend-layout to clarify the complementary scope alongside the existing frontend-design skill
- Added
argument-hint and user-invocable: true frontmatter fields
- Minor formatting normalisation (table alignment, bullet indentation)
License
This project is licensed under MIT.
Portions of the skill definitions are adapted from claude-plugins-official by Anthropic and from openai/skills by OpenAI, both licensed under the Apache License 2.0. Changes were made to adapt the content to VS Code Copilot agent skills format.
