Coding Agent Hosts Verified for Distilly Support: The 8 Compatible Platforms

Distilly natively supports eight verified coding agent hosts—Claude Code, OpenClaw, Hermes, Codex, DeepSeek Harness, Pi, Grok Build, and OpenCode—each with specific invocation commands documented in the repository's SKILL.md file.

Distilly is an open-source skill framework designed to run on any local agent capable of reading files and executing Bash or Python commands. According to the titanwings/distilly source code, these eight platforms have been explicitly tested and verified to discover and invoke Distilly capabilities using standardized slash commands or native UI integration. The compatibility matrix and installation helpers are maintained in the dot-skill/ directory of the repository.

Complete List of Verified Coding Agent Hosts

The dot-skill/SKILL.md file enumerates eight hosts that have passed verification testing. Each platform implements a distinct invocation pattern to activate the Distilly skill:

  • Claude Code: Invoke via /distilly
  • OpenClaw: Use /distilly or fallback to /skill distilly if the native slash command is not registered
  • Hermes: Standard /distilly command
  • Codex: Trigger via $distilly variable syntax or select through the /skills UI
  • DeepSeek Harness: Standard /distilly command
  • Pi coding agent: Requires /skill:distilly syntax
  • Grok Build: Standard /distilly command
  • OpenCode: Uses native Skill discovery automatically without requiring a special command

These definitions are stored in dot-skill/SKILL.md between lines 84-92, which serves as the authoritative compatibility reference for the project.

Host-Specific Installation Paths

The dot-skill/tools/install_generated_skill.py script implements host-aware installation by mapping each verified platform to its default skill directory. Lines 14-22 of this script define the host-to-path mappings, enabling automated deployment across different agent environments.

For example, the script resolves Hermes installations to ~/.hermes/skills/distilly-generated, while other hosts follow similar platform-specific conventions. This abstraction allows developers to install skills using a single command regardless of the target agent.

Installing a Skill for a Specific Host


# Example: install a skill for the Hermes host

python3 tools/install_generated_skill.py \
    --skill-dir "./skills/celebrity/karpathy" \
    --host hermes \
    --force

The --host parameter accepts any of the eight verified identifiers, automatically resolving the correct destination path and printing the final installation location and command name.

How to Invoke Distilly from Each Host

Invocation patterns vary by platform based on their command parsing architectures. The following examples demonstrate the exact syntax required for each verified host:


# Claude Code

/distilly

# OpenClaw

/distilly

# Alternative if slash isn't registered:

/skill distilly

# Hermes

/distilly

# Codex

$distilly

# Or select via the /skills UI

# DeepSeek Harness

/distilly

# Pi coding agent

/skill:distilly

# Grok Build

/distilly

# OpenCode

/distilly  # Automatically discovered, no extra command needed

These patterns are documented in dot-skill/SKILL.md immediately following the host compatibility table. OpenCode differs from other hosts by implementing native Skill discovery, meaning it automatically detects available Distilly capabilities without requiring explicit slash command registration.

Key Source Files Supporting Host Verification

Several files in the dot-skill/ directory define the architecture that enables cross-host compatibility:

Summary

  • Distilly officially supports eight coding agent hosts: Claude Code, OpenClaw, Hermes, Codex, DeepSeek Harness, Pi, Grok Build, and OpenCode.
  • Invocation methods vary by platform, using /distilly, /skill:distilly, $distilly, or native discovery.
  • The install_generated_skill.py script automates host-specific installation using predefined directory mappings.
  • Compatibility definitions reside in dot-skill/SKILL.md, while installation logic lives in dot-skill/tools/install_generated_skill.py.
  • OpenCode is unique among verified hosts for its automatic skill discovery mechanism.

Frequently Asked Questions

What is Distilly and how does it work with coding agents?

Distilly is a skill framework that enables AI coding agents to execute structured tasks by reading files and running Bash or Python commands. It functions as a middleware layer that agents invoke through standardized slash commands or UI elements, allowing the agent to access specialized capabilities without reimplementing core logic.

Can I use Distilly with a coding agent not listed in the verified hosts?

Yes. Distilly is designed to run on any local agent capable of reading files and executing shell commands. While only eight hosts are explicitly verified in SKILL.md, the architecture is host-agnostic. You can manually configure the skill directory and invocation method for unlisted agents using the same installation patterns defined in install_generated_skill.py.

Where are Distilly skills stored on my system for each host?

Storage locations vary by host and are defined in dot-skill/tools/install_generated_skill.py lines 14-22. For example, Hermes uses ~/.hermes/skills/distilly-generated, while other platforms use their own dot-directory conventions within the user's home folder. The installation script automatically resolves these paths based on the --host parameter.

Why does OpenCode not require a special slash command like other hosts?

OpenCode implements native Skill discovery, which automatically detects available capabilities without requiring explicit command registration. While other hosts need specific invocation triggers like /distilly or $distilly, OpenCode discovers Distilly automatically through its native skill inspection mechanism, making integration seamless for that platform.

Have a question about this repo?

These articles cover the highlights, but your codebase questions are specific. Give your agent direct access to the source. Share this with your agent to get started:

Share the following with your agent to get started:
curl -s "https://instagit.com/install.md"

Works with
Claude Codex Cursor VS Code OpenClaw Any MCP Client

Maintain an open-source project? Get it listed too →