Port Skill

/port-skill — Cross-Platform Skill Porter

Purpose: Take a skill from one platform and create its counterpart on the other, applying the correct file format, frontmatter, and conventions for the target platform.

Arguments

• <source-path> — path to the source skill file (required)
• --to claude|copilot — target platform (inferred from source if not specified)

Workflow

1. Detect source platform

Read the source file and determine its platform from the path:

• claude/*/SKILL.md → Claude skill
• copilot-native/prompts/**/*.prompt.md → Copilot prompt
• copilot-native/agents/*.agent.md → Copilot agent
• copilot-native/skills/*/SKILL.md → Copilot skill (knowledge)

If --to is not specified, infer the target as the opposite platform.

2. Read source and extract content

Read the source file. Extract:

• Name/ID from frontmatter
• Description from frontmatter
• Purpose from the first paragraph
• Workflow/steps from the body
• Rules/constraints from any rules section
• References to other skills or assets

3. Map the asset type

If the mapping is ambiguous, ask the user which target type to use.

4. Apply platform conventions

Claude → Copilot-native:

• Replace name: + description: frontmatter with description: (Copilot prompt format)
• Add argument-hint: if the skill has arguments
• Add agent: and tools: if the skill dispatches subagents or uses specific tools
• Replace Claude-specific references:
  • superpowers:* → describe the behavior inline (Copilot has no equivalent)
  • Agent tool → @agent or platform delegation
  • TaskCreate/TaskUpdate → remove or replace with prose
  • CLAUDE.md → copilot-instructions.md or .github/copilot-instructions.md
• Strip Claude Code-specific tool names (Read, Write, Edit, Glob, Grep) — Copilot uses read, search, edit, codebase
• Ask which prompt category: ba/, dev/, review/, common/

Copilot-native → Claude:

• Replace description: frontmatter with name: + description: (Claude skill format)
• Replace tools: list with prose about which Claude Code tools to use
• Replace @agent delegation with Agent tool dispatch
• Replace copilot-instructions.md references → CLAUDE.md
• Replace Copilot tool names (read, search, edit, codebase) with Claude equivalents (Read, Grep, Edit, Glob)
• Strip any applyTo: or glob: frontmatter (Claude uses skill invocation, not auto-apply)

5. Strip environment bake-in

Critical rule from the parity plan: Canonical skills never hardcode environmental specifics. During porting, strip:

• Specific shell names (PowerShell, WSL, bash) — replace with "detect the shell" or reference env.config.md
• Specific hostnames or machine names
• Specific project names, paths, or build commands
• Specific tool choices (npm/yarn/pnpm) — replace with "detect from project config"

If the source has environment-specific content that can't be generalized, note it as a "user must configure" item.

6. Create the target file

Write the ported file to the correct location. Show the full file to the user for review before writing.

7. Update indexes

• If porting to copilot-native and asset-catalog.md exists, add the new file
• If porting to claude, suggest the skill-help category
• If the source references other skills, note which cross-references need porting too

8. Confirm

Print:

Ported: <source-path>
    →   <target-path>

Platform adjustments:
  - Replaced N tool/platform references
  - Stripped M environment-specific items
  - <any warnings about manual follow-up>

Cross-references that may need porting:
  - <referenced skill 1>
  - <referenced skill 2>

Rules

• Don't blindly copy. The port must read naturally on the target platform. A Claude skill that says "use the Agent tool with subagent_type=Explore" should become a Copilot prompt that says "use @agent to explore the codebase."
• Preserve intent, not mechanics. The workflow steps should achieve the same goal using the target platform's idioms.
• Flag what can't be ported. Some Claude features (confidence scores, verdict JSON, Agent tool dispatch) have no Copilot equivalent. Note these as "not ported — target platform handles this differently."
• One skill at a time. Don't batch-port. Each port is a conversation about what translates and what doesn't.
• Environment-neutral. The parity plan's env-neutral principle applies to all ports. No hardcoded shells, hosts, or paths.