Lissom Skills

English · 简体中文 · 日本語

┌─┐
│L│░ LISSOM  —  Simple, reliable Claude Code skills & agents
└─┘  SKILLS     for daily dev automation and context protection.

Why? What's the difference from GSD, SuperPower?

• Zero Dependency — just plain files.
• Thin Skill Dispatchers — relentless context protection.
• Idempotency — hussle-free resume with minimal state.
• Hammered Specs — no surprise dev experience.

/gsd-autonomous	/lissom-auto

When to use?

• I have a scoped idea, help me refine the specs and automate the implementation.

When not to use?

• Trivial or small tasks.
• Exploratory tasks — use /explore.

Basic Workflow

┌─ interview ─┐
           │             /
 research ─┘ auto ──►   +   ──► plan ──► impl ──► review ──► done
  Specs.md    Research.md /    Plan.md         Review.md     │
   ▲                     /                                   │ critical?
   │                     └──────────── fix cycle (max 3)  ◄──┘
   │                                          │
   └──────────────── fix cycles exhausted ────┘

Installation

Install into your project's directory with:

curl -fsSL https://raw.githubusercontent.com/cuzfrog/lissom-skills/main/scripts/install.sh | bash

Supported:

• .claude/ Claude Code and compatible agents.
• .opencode/ OpenCode.
• .qwen/ Qwen Code.
• .gemini/ Gemini CLI.

Uninstallation

Remove all installed files from the current project:

curl -fsSL https://raw.githubusercontent.com/cuzfrog/lissom-skills/main/scripts/uninstall.sh | bash

Only files originally installed by this bundle are removed — any custom files you added are left untouched. Empty directories are cleaned up automatically.

Here We Go!

Run /lissom-auto <task_id> — get interviewed and wait for the job done!

• It looks for the task in .lissom/tasks/<task_id>/Specs.md
• If not found, it tries to locate with tools (e.g. JIRA MCP)

Best practices

• Don't outsource your thinking. Review architecture and code structure. You should understand your code.
• Use a simple grill-me skill to build the Specs.md. (mattpocock/skills)
• Steer the behavior: /lissom-auto <task_id> Directly go to plan phase, I have a well-written spec already.
• Define test methods (development cycle) clearly (e.g. in CLAUDE.md)

Configuration

Set preferences in .lissom/settings.local.json to avoid being asked each run:

{
  "user_attention": "default",
  "fix_threshold": "warning",
  "specreviewrequired": false,
  "impl_delegation": "single"
}

| Key | Options | |---|---| | user_attention | default — Interview for major concerns; auto — Best effort auto pilot; focused — Exhaustive questioning | | fix_threshold | warning — Fix critical & warnings; critical — Critical only; suggestion — All issues | | specreviewrequired | false — Skip spec review; true — Review and refine specs before research | | research_required | true — Perform research before implementation; false — Skip research if the specs are already sufficient | | impl_delegation | single — Perform impl in one subagent (faster in most cases); multiple — Perform impl in multiple subagents (if impl scope is large; usually not necessary) |

Links

• GitHub — source code and releases
• Issues — bug reports and feature requests
• License — MIT

Author

Cause Chung