Get Started

Harness compatibility

The 19 adapters, their install paths, support types and verification tiers.

Skillporter ships adapters for 19 AI agent harnesses. Each adapter records the exact directories that harness reads, how it supports the Agent Skills format, and what evidence backs the claim.

Two questions are kept separate on purpose:

  • Support type describes how the harness consumes a skill.
  • Verification tier describes how the adapter was tested.

A harness can support skills natively and still only be fixture tested. Skillporter never marks an adapter as runtime verified because a directory was created.

Support types

Support type Meaning
NATIVE The harness reads the Agent Skills format (SKILL.md) directly.
COMPATIBLE The harness reads SKILL.md through a documented compatibility path.
ADAPTED The harness needs a controlled conversion, such as a rules or conventions file.

The registry also defines PARTIAL, UNSUPPORTED and UNVERIFIED for custom adapters. No built-in adapter uses them.

Verification tiers

Tier Evidence required
RUNTIME_VERIFIED Installation and live discovery verified inside the running harness on a test host.
INTEGRATION_TESTED End-to-end multi-harness workflows (add, sync, diagnose, doctor --fix, drift, duplicates, shared-path deduplication, rollback) verified in automated integration suites, plus vendor documentation.
FIXTURE_TESTED The full adapter lifecycle verified against automated fixtures in tests/test_adapter_conformance.py, plus official vendor documentation.
DOCUMENTATION_VERIFIED Discovery paths and precedence confirmed against vendor documentation, without automated fixture execution.

Custom adapters loaded with load_adapter_file() default to UNVERIFIED until you validate them yourself.

The 19 adapters

This table is generated from skillporter/harnesses.py, the same registry the CLI uses. Do not edit it by hand. Run python3 website/scripts/generate_site_data.py after changing the registry.

ID Harness Category Support Verification Project path User path Vendor docs
hermes Hermes Agent autonomous NATIVE RUNTIME_VERIFIED .hermes/skills $HERMES_HOME/skills Docs
claude-code Claude Code coding NATIVE INTEGRATION_TESTED .claude/skills ~/.claude/skills Docs
codex OpenAI Codex coding NATIVE INTEGRATION_TESTED .agents/skills ~/.agents/skills Docs
cursor Cursor coding NATIVE INTEGRATION_TESTED .agents/skills ~/.agents/skills Docs
github-copilot GitHub Copilot coding NATIVE FIXTURE_TESTED .github/skills ~/.copilot/skills Docs
gemini-cli Gemini CLI coding NATIVE FIXTURE_TESTED .agents/skills ~/.agents/skills Docs
antigravity Google Antigravity coding NATIVE FIXTURE_TESTED .agents/skills ~/.gemini/config/skills Docs
windsurf Windsurf coding NATIVE FIXTURE_TESTED .windsurf/skills ~/.codeium/windsurf/skills Docs
cline Cline coding NATIVE FIXTURE_TESTED .cline/skills ~/.cline/skills Docs
roo-code Roo Code coding NATIVE FIXTURE_TESTED .roo/skills ~/.roo/skills Docs
opencode OpenCode coding NATIVE FIXTURE_TESTED .opencode/skills ~/.config/opencode/skills Docs
zed Zed coding NATIVE FIXTURE_TESTED .agents/skills ~/.agents/skills Docs
replit Replit Agent coding NATIVE FIXTURE_TESTED .agents/skills ~/.config/agents/skills Docs
junie JetBrains Junie / AI Assistant coding NATIVE FIXTURE_TESTED .junie/skills ~/.junie/skills Docs
openclaw OpenClaw autonomous NATIVE FIXTURE_TESTED .openclaw/skills ~/.openclaw/skills Docs
openhands OpenHands autonomous NATIVE FIXTURE_TESTED .agents/skills ~/.agents/skills Docs
goose Goose autonomous NATIVE FIXTURE_TESTED .agents/skills ~/.agents/skills Docs
continue Continue coding COMPATIBLE FIXTURE_TESTED .continue/skills ~/.continue/skills Docs
aider Aider / AiderDesk coding ADAPTED FIXTURE_TESTED .aider-desk/skills ~/.aider-desk/skills Docs

19 adapters: 1 RUNTIME_VERIFIED, 3 INTEGRATION_TESTED, 15 FIXTURE_TESTED.

Print the same matrix locally:

skillporter matrix
skillporter matrix --category autonomous
skillporter matrix --json

Inspect one adapter, including every compatibility path and known limitation:

skillporter info cursor

The shared .agents/skills directory

Eight harnesses use .agents/skills as their primary project directory: Codex, Cursor, Gemini CLI, Antigravity, Zed, Replit Agent, OpenHands and Goose. Six more read it as a documented compatibility path: GitHub Copilot, Cline, Roo Code, OpenCode, Junie and OpenClaw.

Skillporter treats that directory as shared property.

  • One write. skillporter add <source> --harness all writes .agents/skills/<skill> once and reports the other owners as sharing the path.
  • Safe removal. skillporter remove <skill> --harness cursor removes any Cursor-only copy but keeps .agents/skills/<skill> while another managed harness still owns it. The shared copy is deleted only when its last owner is removed.
  • No double loading. Installing the same skill into .agents/skills and .cursor/skills makes Cursor list it twice. skillporter diagnose reports that as DUPLICATE.

Harness differences that matter

Harnesses do not behave identically, and Skillporter does not pretend they do. A few examples from the registry:

  • Claude Code reads .claude/skills only. A skill that exists only in .agents/skills is invisible to it.
  • Codex and Zed read .agents/skills only.
  • Gemini CLI requires the project folder to be trusted before it loads workspace skills.
  • Cline expects the SKILL.md body to stay under 5,000 tokens.
  • Core Aider does not discover skill folders on its own. It needs aider --read <skill>/SKILL.md or an adapted CONVENTIONS.md.

skillporter info <harness> prints the notes and limitations for any adapter.

Detecting what you have installed

skillporter detect .
skillporter detect . --all --versions

Detection is read-only. It looks for known binaries, user directories, project markers and environment variables.

Adding your own harness

See Custom adapters for the declarative JSON format and the Python registration API.

Type to search the documentation.