Hermes v2.0.0-beta — Task-First Natural Language Interface

Hermes v2.0.0-beta — Task-First Workflow

Release: v2.0.0-beta · Date: 2026-06-20
⚠️ Beta / Experimental — Not production-ready. v1.1.0 remains stable.

---

What is Hermes v2 Beta?

Hermes v2 adds a natural-language frontend to the Hermes Agent skill system. Instead of writing constraint-engineering prompts by hand, users say what they want in plain language.

v1:  User writes "## Authority\n你是 Release Manager\n## Mission\n..."
v2:  User says "帮我发布项目" → system handles everything automatically

---

Why Task-First?

v1 exposed internal concepts (roles, skills, constraints) directly to users. This created friction:

• "Which role do I choose?" (8 personas)
• "Which skill do I load?" (8 SKILL.md files)
• "How do I write a constraint prompt?" (Authority, Mission, Constraints...)

v2 inverts the model: the user states the goal; the system figures out how.

---

Features

Three User Modes

Mode	Input	Behavior
Beginner	"帮我发布项目"	Full auto: detect intent → select skills → generate constraints → execute
Advanced	"用 Release Manager 发布"	User picks persona; system handles the rest
Expert	## Authority\n你是...	Pure v1 — bypasses v2 pipeline entirely

Entity Extraction

v2 recognizes specifics in user input and injects them into the execution plan:

• Technology: "FastAPI", "Django", "React" → constraint prompt includes tech stack
• File path: "test_router.py" → debug workflow targets the right file
• Version: "v2.0.0" → release workflow uses the exact version
• Document type: "README" → docs workflow knows what to write

Confidence Calibration

Every routing decision includes a confidence score (0–100%). Below 40% triggers a clarification question. Real-world dogfood inputs average 71.3% confidence.

6 Built-in Tasks

Task	Trigger (CN)	Trigger (EN)	Skills
publish_project	帮我发布项目	publish the project	cicd-orchestrator, code-quality-guardian
fix_bug	修复这个错误	fix the bug	debugger-coordinator, code-quality-guardian
create_project	创建一个项目	create a project	requirement-analyzer, spec-driven-dev, test-driven-dev
write_docs	写文档	write docs	(auto-detect)
review_code	检查代码	review code	code-quality-guardian
release_version	发布 v1.2.0	release v1.2.0	cicd-orchestrator, code-quality-guardian

---

Backward Compatibility

v1 is unchanged. All existing functionality works as before:

# v1 CLI — always available
hermes-skill validate skills/
hermes-skill create my-skill
hermes-skill list skills/

# v2 Beta — new entry point
python -m hermes_v2.cli "帮我发布项目"

• 8 SKILL.md files: unmodified
• hermes-skill CLI: unchanged
• v1 tests (104): all passing
• v2 is in src/hermes_v2/ + tasks/ — no v1 files touched

---

Known Limitations

Limitation	Impact	Plan
Rule-based Router	May misfire on ambiguous inputs	LLM Router in Phase 2
6 task types	Limited coverage	Expand based on user feedback
No Persian/Farsi/etc.	CN + EN only	Language contributions welcome
Entity extraction is regex-based	Misses novel tech names	Expand knowledge base over time
CLI display is minimal	No rich output formatting	Improve in v2.1

---

Try It

# Clone (if not already)
git clone https://github.com/Ow1onp/hermes-agent-skills.git
cd hermes-agent-skills

# Install (v1 way — still works)
pip install -e .

# v2 Beta — natural language
python -m hermes_v2.cli "创建一个 FastAPI 项目"
python -m hermes_v2.cli "帮我发布项目" --dry-run
python -m hermes_v2.cli "修复这个错误" --verbose

# Run all tests
python -m pytest tests/ -q

---

Rollback

v2 is purely additive. To remove it:

# Option 1: Just don't use v2
# Keep using hermes-skill as before

# Option 2: Remove v2 files
rm -rf src/hermes_v2/ tasks/ tests/test_hermes_v2*.py docs/v2/

# Option 3: Stay on v1 tag
git checkout v1.1.0

Zero data loss. Zero downtime. Zero v1 impact.

---

What's Next

• Your feedback → Open an Issue or Discussion
• New task types → Submit a Skill Request
• Language support → PRs welcome for new languages
• v2.0.0 stable → After Beta feedback cycle (4–6 weeks)

---

This is Beta software. APIs may change. Report issues on GitHub.

v1.1.0 — The CLI Toolchain Release

🧬 hermes-agent-skills v1.1.0 — The CLI Toolchain Release

Ship skills like you ship code. Validate, scaffold, and audit — all from the terminal.

---

Release Summary

v1.1.0 transforms hermes-agent-skills from a curated skill collection into a full developer toolchain. The centerpiece is hermes-skill — a Typer-powered CLI that brings the same rigor to skill authoring that you already apply to code. Scaffold standards-compliant SKILL.md files from templates, validate them against the Agent Skills Open Standard with 6-dimension analysis, and audit entire skill repositories with a single command.

Under the hood, the SkillValidator received its most significant upgrade yet, growing from 2 to 6 validation dimensions with configurable strictness, SemVer checking, reserved-name detection, and recursive directory scanning. The test suite has more than doubled — 104 tests, all passing, covering every CLI path and validator edge case.

And because great tooling deserves great documentation: v1.1.0 ships with a full 5-document documentation suite — Quick Start, Tutorial, FAQ, Contributing Guide, and an overhauled bilingual README — so your first skill is 10 minutes away, not an afternoon of reading specs.

---

✨ Highlights

hermes-skill CLI	Create, validate, and list skills from the terminal — Typer-native, --help everywhere
6-Dimension Validator	Frontmatter syntax, SemVer compliance, trigger quality, name-directory consistency, reserved name detection, license metadata
Scaffolding Templates	basic, advanced, minimal — pick the right starting point for any skill
104 Tests (↑126%)	Full CLI integration coverage + validator edge cases, all passing
Documentation Suite	Quick Start, Tutorial, FAQ, and Contributing Guide — engineered for progressive disclosure

---

Added

hermes-skill CLI

Command	Description
hermes-skill create <name>	Scaffold a new SKILL.md from templates (basic / advanced / minimal) with --category and --no-interactive flags
hermes-skill validate <path>	Validate one or more skills with --strict (require all recommended fields), --quiet (only failures), and --recursive (scan directories)
hermes-skill list [path]	List discovered skills with --format table or --format json, featuring status icons, dimensions, and scores
hermes-skill soul generate	Generate SOUL.md persona templates (balanced / architect / pragmatist)
hermes-skill soul read	Read and parse SOUL.md files into structured output

Enhanced SkillValidator (v2)

• 6 validation dimensions with individual pass/fail/warn results:
  1. Frontmatter syntax (--- delimiters, YAML parsing)
  2. Required fields (name, description, version)
  3. Field constraints (name format regex, description ≤1024 chars)
  4. Semantic validity (SemVer compliance, lowercase-hyphen names)
  5. Recommended quality (≥3 triggers, resolution checklist, metadata block)
  6. Structural integrity (name-directory match, reserved name detection)
• --strict mode requiring all recommended quality fields
• --quiet mode showing only failures
• --recursive flag for directory-wide audit
• Reserved name detection (claude, anthropic, openai, etc.)
• License and compatibility metadata validation

Documentation (docs/)

Document	Description
QUICKSTART.md	10-minute guide — install, create your first skill, validate, iterate
TUTORIAL.md	Deep dive into SKILL.md anatomy, advanced patterns, and integration tricks
FAQ.md	40+ real-world Q&As grouped by user intent (getting started, authoring, troubleshooting)
CONTRIBUTING.md	How to contribute skills, code, docs, or templates — with style guide and PR checklist

Tests

• 58 new tests bringing the suite from 46 → 104 (↑126%)
• CLI integration tests for create, validate, and list (tests/test_cli/)
• Validator edge-case tests: SemVer parsing, trigger validation, reserved names, metadata constraints
• All 104 tests pass on Python 3.10–3.12

---

Changed

• SkillValidator — Refactored from 2D to 6D validation framework with configurable strictness and structured result objects
• pyproject.toml — cli package declared as installable; entry point hermes-skill = cli.main:run
• README.md — Bilingual split: Chinese (primary) + English (README.en.md showcase) with modern aesthetic

Fixed

• console_scripts entry point — Corrected to hermes-skill (was hermes-skills in v1.0.0 scaffolding)

---

Documentation

Resource	Link
📖 README (中文)	README.md
📖 README (English)	README.en.md
⚡ Quick Start	docs/QUICKSTART.md
🎓 Tutorial	docs/TUTORIAL.md
❓ FAQ	docs/FAQ.md
🤝 Contributing	CONTRIBUTING.md
📋 Changelog	CHANGELOG.md

---

Installation

pip (recommended)

pip install hermes-agent-skills==1.1.0

From source

git clone https://github.com/Ow1onp/hermes-agent-skills.git
cd hermes-agent-skills
git checkout v1.1.0
pip install -e ".[dev]"

Verify

hermes-skill --help
hermes-skill validate . --recursive --quiet

---

Compatibility

	Minimum	Recommended
Python	3.10	3.11+
Operating System	Linux, macOS, Windows	—
Hermes Agent	Any version with /skills support	Latest release
Agent Skills Standard	v1.0.0	v1.0.0

Dependencies

• pyyaml >= 6.0
• typer >= 0.9
• No system-level dependencies required

---

Future Roadmap

The project is converging toward a Skill Analytics SaaS — think "npm audit for AI agent skills." The CLI toolchain in v1.1.0 is the first building block.

v1.2.0 — Analytics & Quality Scoring

• hermes-skill score — multi-axis quality scoring with letter grades
• hermes-skill diff — semantic diff between skill versions
• hermes-skill stats — aggregate metrics across skill collections
• Pre-commit hook integration (hermes-skill validate as a git hook)

v1.3.0 — Marketplace & Discovery

• hermes-skill search <query> — discover community skills
• hermes-skill install <name> — one-command skill installation
• Skill registry with version resolution
• Compatibility matrix (skill × Hermes version × OS)

v2.0.0 — SaaS Platform

• Web dashboard with real-time skill health monitoring
• Automatic evolution suggestions via LLM analysis
• Team collaboration — shared skill collections with permissions
• Enterprise SSO and audit logging

---

Full Changelog: v1.0.0...v1.1.0

---

_{Built with ❤️ by Ow1onp · MIT · Agent Skills Standard}

Full Changelog: v1.0.0...v1.1.0