From clamper
One-command cross-platform agent ecosystem initialization. Analyzes codebase DNA and scaffolds CLAUDE.md, AGENTS.md, subagents, skills, and memory — tailored to the actual project.
How this skill is triggered — by the user, by Claude, or both
Slash command
/clamper:ecosystem-initThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Most CLAUDE.md files are hand-written, generic, and go stale. Most agent ecosystems are either missing or copy-pasted from templates. This skill analyzes the actual codebase and generates an ecosystem that reflects the real project — not a template.
Most CLAUDE.md files are hand-written, generic, and go stale. Most agent ecosystems are either missing or copy-pasted from templates. This skill analyzes the actual codebase and generates an ecosystem that reflects the real project — not a template.
The full cross-platform ecosystem structure:
<project-root>/
│
│ ── Cross-Platform Layer (works with ALL agent tools) ──
├── CLAUDE.md # Claude Code native rules (the source of truth)
├── AGENTS.md → CLAUDE.md # Symlink: Codex, Gemini, Copilot, Cursor all read this
├── .agents/
│ └── skills/ # Open standard skill format
│ └── <project>/SKILL.md # Project-specific workflow skill
│
│ ── Claude Code Layer (richest implementation) ──
├── .claude/
│ ├── agents/ # Subagents tailored to this project
│ │ ├── code-reviewer.md # Knows this project's conventions
│ │ ├── test-writer.md # Knows this project's test framework
│ │ └── <conditional>.md # Stack-specific agents
│ ├── skills/ → ../.agents/skills/ # Points to cross-platform skills
│ └── memory/
│ ├── MEMORY.md # Index
│ └── project-dna.md # Full DNA extraction results
│
│ ── Cursor Compatibility (if applicable) ──
├── .cursor/
│ └── skills/ → ../.agents/skills/
│
│ ── Tool Config (if applicable) ──
└── .mcp.json # MCP server config
| Generated File | Claude Code | Codex | Gemini CLI | Cursor | Copilot |
|---|---|---|---|---|---|
| CLAUDE.md | Native | — | — | Reads it | — |
| AGENTS.md (symlink) | — | Native | Reads it | Reads it | Reads it |
| .agents/skills/*/SKILL.md | Via symlink | Native | Native | Via symlink | Native |
| .claude/agents/*.md | Native | — | — | Can read | — |
| .claude/memory/ | Native | — | — | — | — |
Key insight: By placing skills in .agents/skills/ and symlinking from .claude/skills/ and .cursor/skills/, the same skill works on 5+ platforms with zero duplication.
When the project has code (>3 source files or a manifest file):
When the project is empty or nearly empty:
A generated CLAUDE.md must meet these criteria:
Bad CLAUDE.md (generic):
## Rules
- Write clean code
- Add tests
- Follow best practices
Good CLAUDE.md (specific):
## Rules
- Use pytest-django with factory_boy fixtures (see apps/core/tests/factories.py for patterns)
- All views in apps/*/views.py must inherit from LoginRequiredMixin
- Templates use {% block content %} and {% include "partials/_*.html" %} pattern
- HTMX attributes go on the triggering element, not the target
Generated subagents must:
/src/tests/)haiku model for fast tasks (review), sonnet for complex tasks (test writing)The /init command should suggest:
/clamp to verify the current codebase state/dna for deeper architecture analysis beyond what /init capturednpx claudepluginhub pretinnov-inc/claude-plugin-marketplace --plugin clamperProvides a checklist for code reviews covering functionality, security, performance, maintainability, tests, and quality. Use for pull requests, audits, team standards, and developer training.