Skip to content

Repository files navigation

Hermes SkillForge — Universal Agent Skill Converter

CI npm version npm downloads Python License: MIT Hermes Agent Skill Agent Skills Standard

Convert any skill, repository, or tool into an install-ready Hermes Agent skill.

Agent Skills → Claude Skills → Codex → Cursor Rules → MCP Servers → Prompt Packs → GitHub Repos → Plugins → Hermes Skills

Installation · Quick Start · Features · Docs · Contributing · License


What is Hermes SkillForge?

Hermes SkillForge is an open-source, zero-dependency conversion engine that transforms diverse AI-agent source formats into valid, install-ready Hermes Agent skills that follow the official authoring standards and pass automated validation and testing.

Whatever you have — an agentskills.io skill, a Claude or Codex skill, Cursor rules, an MCP server manifest, a GitHub repository containing an agent workflow, a plugin, a prompt pack, or a script collection — SkillForge analyzes it, maps its capabilities, generates a compliant SKILL.md, builds a per-skill test suite, and installs it into ~/.hermes/skills/.

Live demo (real output)

$ skillforge convert ./examples/agentskills-source

== Source analysis ==
Format: agent_skills (confidence: medium, score: 2.0)
  - SKILL.md at SKILL.md
  - 1 Python file(s) with function definitions
Capabilities:
  commands: 0, dependencies: 0, scripts: 1, API endpoints: 0, config keys: 1

== Conversion complete ==
Generated: /tmp/hermes-demo1/skillforge/text-transform-pipeline
Dirs created: ['scripts/', 'references/', 'tests/']

== Validation ==
VALID

== Tests ==
passed: 18/18

Action: fresh -> ~/.hermes/skills/software-development/text-transform-pipeline
Installed. Active in a new session; run /reset for the current one.

Features

Core capabilities

Feature Description
Universal format detection Scored multi-evidence classifier recognizes 20+ formats including SKILL.md, mcp-server.json, plugin.yaml, .mdc Cursor rules, *.mcp.json, and GitHub repo conventions
Official Hermes SKILL.md generation Modern section order (When to Use → Procedure → Verification), frontmatter, and the 60-character description budget enforced automatically
Per-skill test generation Every converted skill ships a tests/ pytest suite covering structure, scripts, dependencies, invalid inputs, and missing files
0–100 quality scoring Validation state, test pass rate, frontmatter completeness, and description budget compliance are scored numerically
Auto-repair engine Iterates targeted fixes (name collisions, frontmatter fences, missing references, placeholder scripts) with a full attempt history and blocker report
Secret detection & redaction API keys, tokens, and passwords are detected, redacted from output, and reported in references/security-findings.md
Safe placeholder handling Host-dependent scripts (rsync, npm, kubectl, GUI tools) are statically validated instead of executed, so conversion never fails on unrunnable workflows
MCP-to-skill conversion Tools are extracted from manifests and surfaced in the skill's Procedure; secret-backed config keys become metadata.hermes.config prompts
Zero dependencies The engine is pure Python 3.10+ stdlib — no pip install required, no network calls at runtime
Rollback & conflict safety Installs snapshot previous versions and detect naming conflicts before overwriting

Supported source formats

Category Formats
Agent skill standards agentskills.io SKILL.md, Claude Skills, Codex skills, Cursor rules (.mdc), .claude/.codex/.agents conventions
Repository sources GitHub repos with agent workflows, plugin repositories (plugin.yaml, register()), script collections
Integrations MCP server manifests (mcp-server.json, *.mcp.json, mcpServers/tools JSON), REST API specs
Knowledge sources Prompt packs, rules files, markdown documentation, YAML/JSON configs
Existing skills Already-valid Hermes skills (re-import, upgrade, or validate)

Architecture

Hermes SkillForge conversion pipeline

The pipeline flows analyze → convert → validate → repair → test → install. The repair stage loops back to validate until the skill is clean or a genuine blocker is reported, and every unmapped capability is preserved in README.md and references/unmapped-functionality.md — nothing is ever silently dropped.

Installation

SkillForge is distributed on npm under hermes-skillforge and can be installed with any major JavaScript/Node toolchain. Python ≥ 3.10 is the only runtime requirement (the engine itself is stdlib-only).

npm

npm install -g hermes-skillforge
skillforge --help

Yarn

yarn global add hermes-skillforge
skillforge --help

pnpm

pnpm add -g hermes-skillforge
skillforge --help

Bun

bun add -g hermes-skillforge
skillforge --help
# or run directly from this repo:
bun run main.bun.js convert <source>

Deno

# No install needed — run directly against the repo:
git clone https://github.com/uthumany/hermes-skillforge.git
cd hermes-skillforge
deno run --allow-read --allow-write --allow-env Deno.ts convert <source>

npx (no install at all)

npx -y hermes-skillforge convert <source>

Rush

# Add to your monorepo rush.json projects:
#   { "packageName": "hermes-skillforge", "projectFolder": ".", "shouldPublish": true }
rush add -p hermes-skillforge

Lerna

# Inside a Lerna monorepo, add to package.json dependencies:
npm install hermes-skillforge
npx lerna run test --scope hermes-skillforge

Volta (version-pinned)

volta install node
volta install hermes-skillforge
skillforge --help

Install directly from GitHub (Python)

git clone https://github.com/uthumany/hermes-skillforge.git
cd hermes-skillforge/skill
python3 scripts/skillforge.py convert <source>

Quick Start

# Analyze any source (URL, repo, directory, or ZIP)
skillforge analyze <url-or-path>

# Full pipeline: import → analyze → convert
skillforge import <url-or-path>

# Convert only, then validate and test
skillforge convert <source>
skillforge validate
skillforge test

# Preview the generated SKILL.md
skillforge preview

# Install into ~/.hermes/skills/<category>/
skillforge install

# Convert every convertible source under a directory
skillforge batch ./my-sources/

# Rolled back? Restore a previous install
skillforge rollback

Every converted skill lands in ~/.hermes/skills/<category>/ and becomes active in the next Hermes Agent session (run /reset for the current one).

Usage Examples

Example 1 — Convert an agentskills.io skill

skillforge convert ./examples/agentskills-source
skillforge validate            # VALID
skillforge test                # passed: 18/18
skillforge install             # -> ~/.hermes/skills/software-development/text-transform-pipeline

Example 2 — Convert a GitHub repo with an agent workflow

git clone <agent-workflow-repo>
skillforge convert ./agent-workflow-repo
# Detected: shell scripts + workflow README
# Generated SKILL.md preserves review.sh / post-comment.sh,
# redacts GH_TOKEN, and lists the remote deploy steps as references

Example 3 — Convert an MCP server into a Hermes skill

skillforge convert ./examples/mcp-source
# Detected: mcp_server (high confidence, score 4.5)
# tools weather_current / weather_forecast extracted into the Procedure;
# WEATHER_API_KEY surfaced as a config prompt in SKILL.md frontmatter

Folder Structure

hermes-skillforge/
├── bin/skillforge              # CLI entrypoint (npm bin)
├── docs/architecture.d2|svg    # Pipeline architecture diagram
├── examples/                   # Example source fixtures
│   ├── agentskills-source/     # agentskills.io-compatible skill
│   ├── workflow-repo-source/   # GitHub-style repo with agent workflow
│   └── mcp-source/             # MCP server manifest (mcp-server.json)
├── skill/                      # The installable Hermes skill
│   ├── SKILL.md                # Main skill document
│   ├── scripts/skillforge.py   # Conversion engine (stdlib-only, ~2,150 lines)
│   ├── references/             # Format detection, conversion rules, validator rules,
│   │   │                       # MCP conversion, installation guides
│   ├── templates/              # Packaging scaffolding
│   ├── tests/                  # Engine self-suite (10/10 passing)
│   └── README.md
├── scripts/skillforge          # Shell wrapper
├── Deno.ts                     # Deno runner
├── main.bun.js                 # Bun runner
├── rush.json                   # Rush monorepo manifest
├── package.json                # npm manifest
├── .github/                    # CI workflow, issue/PR templates, funding, security
├── CHANGELOG.md
├── CONTRIBUTING.md
├── CODE_OF_CONDUCT.md
├── SECURITY.md
└── LICENSE                     # MIT

Documentation

Document Contents
CONTRIBUTING.md How to contribute and add new formats
CHANGELOG.md Version history
SECURITY.md Security policy and vulnerability reporting
skill/references/format-detection.md Format scoring and evidence rules
skill/references/conversion-rules.md Capability extraction and mapping rules
skill/references/validator-rules.md Hermes SKILL.md validation rules
skill/references/mcp-conversion.md MCP manifest → skill conversion
skill/references/installation.md Hermes skill installation details
docs/architecture.svg Pipeline architecture diagram

Contributing

Contributions are welcome! Please read CONTRIBUTING.md and our Code of Conduct first. Quick summary: fork → branch → test → PR (use the PR template). Adding a new source format takes roughly five focused changes — the guide walks you through them.

Badges & Keywords

This repository is optimized for discoverability with the following keywords and topics: hermes-agent, hermes-skill, agentskills.io, claude-skills, codex, cursor-rules, mcp-server, mcp, ai-agent, agent-skill, skill-converter, automation, claude, llm-tools, prompt-engineering, devtools.

License

This project is licensed under the MIT License — see LICENSE for details.

Acknowledgements

Built on the Hermes Agent skill architecture and the agentskills.io open standard.

Show your support

If Hermes SkillForge is useful to you, starring the repository helps others find it and directly supports continued development. Bug reports, feature requests, and pull requests are all welcome — this project follows a community-driven roadmap in CONTRIBUTING.md.

About

Convert any skill, repo, or tool into an install-ready Hermes Agent skill. 20+ formats: Agent Skills, Claude, Codex, Cursor, MCP, plugins, prompt packs.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages