Skillport

Note: Run scripts/render-context.sh before first use to inject dynamic context.

skillport: Universal Skill/Rule Adapter

Step 1: Understand the Request

Parse the user's prompt to extract three things:

1. WHAT — Which skill/rule/config to convert
   • A path (e.g., .claude/skills/my-skill, CLAUDE.md, .cursor/rules/)
   • A skill name (e.g., "clearshot", "toc") — resolve to its installed path
   • "all" or "everything" — scan for all detected configs
2. FROM — Source harness format
   • Explicit: "from claude", "from cursor", etc.
   • Implicit: auto-detect from path and file contents
3. TO — Target harness format(s)
   • Explicit: "to cursor", "to codex and openclaw", "to all"
   • One or more of: claude, cursor, codex, openclaw, copilot, windsurf, all

Examples of natural language parsing:

User says	WHAT	FROM	TO
"convert clearshot to cursor"	clearshot	auto-detect	cursor
"port my claude skills to codex"	all claude skills	claude	codex
"make this cursor rule work in claude code"	(ask which rule)	cursor	claude
"adapt toc for all harnesses"	toc	auto-detect	all
"what harness configs do I have?"	(detect mode)	—	—

Resolution Rules:

• If WHAT is a name (not a path), search these locations:

  • ~/.claude/skills/<name>/
  • ~/.agents/skills/<name>/
  • .claude/skills/<name>/ (project-local)
  • .agents/skills/<name>/ (project-local)
  • .cursor/skills/<name>/
  • .cursor/rules/<name>.mdc

• If FROM is not stated, auto-detect by running the CLI:

  npx tsx ~/.claude/skills/skillport/bin/skillport.ts detect .
  

• NEVER start conversion until WHAT, FROM, and TO are all resolved.

Step 2: Confirm Before Converting

Always show the user what you're about to do:

I'll convert:
  Skill:  <name> (<path>)
  From:   <harness> (auto-detected / specified)
  To:     <target harness(es)>

Proceed?

Wait for confirmation. If the user says "just do it" or similar, proceed.

Step 3: Run the Conversion

Execute the CLI:

npx tsx ~/.claude/skills/skillport/bin/skillport.ts convert <source-path> --to <targets> --output <dir>

For dry run (preview without writing):

npx tsx ~/.claude/skills/skillport/bin/skillport.ts convert <source-path> --to <targets> --dry-run

For project scan:

npx tsx ~/.claude/skills/skillport/bin/skillport.ts detect <dir>

Step 4: Present the Result Report

After conversion, ALWAYS present the full report to the user. The report has three mandatory sections:

Section 1: Conversion Summary

Show what was converted and where files were written:

skillport: <name> (<from> → <to>)

  Source:  <source path>
  Target:  <target path>

  Fields:  N total
    ✓  X native   (fields that mapped 1:1)
    ⚡  Y shimmed  (fields with functional approximations)
    ⚠  Z dropped  (fields with no equivalent)

Section 2: Parity Assessment

Show a percentage score and per-feature breakdown:

Parity: NN% (Full|High|Partial|Low)

  Feature Coverage:
    ✓ Core instructions       100%  Markdown body preserved
    ✓ Activation trigger      100%  description-based discovery
    ⚡ Tool restrictions        80%  Instruction text, not enforced
    ⚠ Hooks                     0%  No equivalent in target
    ...

  Verdict: <plain-English assessment>

Parity levels:

• 95-100% Full — skill works identically
• 80-94% High — core behavior preserved, minor features shimmed
• 50-79% Partial — key features approximated, some lost
• <50% Low — significant gaps, manual adaptation recommended

Section 3: Tradeoffs & Key Points

Short bullet list of the most important things to know:

Key Points:
• <what changed and why it matters>
• <what the user should watch out for>
• <round-trip safety notes>

Supported Harnesses

Harness	Skills	Rules	Instructions	Hooks	Subagents	Globs	Dynamic Context
Claude Code	✓ native	✓ native	CLAUDE.md	✓ 24 events	✓ context:fork	✓ paths	✓ !cmd
Cursor	✓ native	✓ .mdc	AGENTS.md	⚡ partial	✓ native (v2.0+)	✓ globs	✗ none
Codex CLI	✓ native	via AGENTS.md	AGENTS.md	✗ none	⚡ fork cmd	✗ none	✗ none
OpenClaw	✓ native	via skill	N/A	⚡ gateway	✗ none	✗ none	⚡ preamble
Copilot	✗ none	✓ .instructions.md	copilot-instructions.md	✗ none	✗ none	✓ applyTo	✗ none
Windsurf	✗ rules only	✓ .windsurfrules	.windsurfrules	✗ none	✗ none	✗ none	✗ none

Feature Adaptation Quick Reference

• Hooks: CC→Cursor maps PreToolUse→beforeShellExecution. Others get wrapper scripts or annotations.
• Subagents: CC→Cursor is near-native. Codex has fork. Others get instruction annotations.
• allowed-tools: Only CC enforces per-skill. Others get instruction text (agent may ignore).
• Globs: CC uses paths:, Cursor uses globs:, Copilot uses applyTo:. Others get annotations.
• Dynamic context (!cmd``): CC-only native. Others get preprocessor scripts or preamble blocks.
• Model/effort overrides: CC-only. Dropped with annotation for others.