Communitygithub.com

grimoire-rs/grimoire

Package manager for AI-agent config. grim installs, updates, and publishes skills, rules, agents, MCP servers, and bundles into Claude Code, Copilot, Cursor, Codex, Gemini, Zed, Amp, Kiro, Junie, and opencode — pinned by digest in a lockfile. Storage is any OCI registry; there is no service to run.

grimoire 是什麼?

grimoire is a Claude Code agent skill that package manager for AI-agent config. grim installs, updates, and publishes skills, rules, agents, MCP servers, and bundles into Claude Code, Copilot, Cursor, Codex, Gemini, Zed, Amp, Kiro, Junie, and opencode — pinned by digest in a lockfile. Storage is any OCI registry; there is no service to run.

相容平台Claude CodeCodex CLICursorGemini CLIOpenCode
npx skills add grimoire-rs/grimoire

在你喜歡的 AI 中提問

開啟一個已預先載入此 Agent Skill 的新對話。

說明文件

Grimoire Documentation

Role: write user-facing docs for Grimoire (Markdown under docs/).

Workflow

  1. Read source code — no memory docs
  2. Read product context.claude/rules/product-context.md before user-facing writing
  3. Search real-world examples from other ecosystems before comparisons
  4. Identify problem feature solves before solution
  5. Draft narratively — idea → problem → solution → depth
  6. Verify internal links point to existing sections with content

Narrative Standards

  • Idea → problem → solution, then depth — never start "Grimoire is a..."
  • No marketing tone — examples make case
  • Reference-style links only — never inline [text](url); definitions at file bottom
  • Every external tool hyperlinked — every occurrence, not first
  • Analogies in :::info callout boxes, not inline
  • Custom anchors on every heading{#parent-subsection}

Relevant Rules (load explicitly for planning)

  • .claude/rules/docs-style.md — narrative + linking + anchor conventions
  • .claude/rules/product-context.md — positioning and product identity

Tool Preferences

  • WebFetch / WebSearch — real-world examples from other tools before comparisons

Constraints

  • NEVER inline links — reference-style only
  • ALWAYS read source before documenting; never from memory

Handoff

  • To /hex-architect — docs revealing design ambiguity
  • To Builder — code changes uncovered while writing docs

$ARGUMENTS

相關技能