Rosetta Onboard
You are onboarding this repository into rosetta — creating a unified .agents/ directory that serves as the single source of truth for all AI agent configurations across IDEs.
Step 1: Scan the repository
Search for all existing agent infrastructure files. Check for ALL of the following:
Claude Code artifacts:
CLAUDE.md(project root and.claude/CLAUDE.md).claude/rules/*.md.claude/skills/*/SKILL.md.claude/agents/*.md.claude/settings.json(look forpermissions.denypatterns andmcpServers).claude/commands/*.md(legacy commands)
Cursor artifacts:
.cursorrules(deprecated but may exist).cursor/rules/*.mdc.cursor/skills/*/SKILL.md.cursor/agents/*.md.cursor/hooks.json.cursor/commands/*.md.cursorignore.cursorindexingignore.mcp.json
Cross-IDE artifacts:
.agents/(already exists — rosetta may already be partially set up)AGENTS.md
List everything you find with a brief description of each file's purpose and content.
Step 2: Analyze and propose
Based on what you found, propose a .agents/ directory structure. Consider:
Rules: Consolidate overlapping rules from
.cursor/rules/and.claude/rules/. If a Cursor rule and a Claude Code rule cover the same topic, merge them into one canonical rule. Usepaths:(notglobs:) for path scoping in canonical format.Skills: Skills use the Agent Skills standard and are identical across IDEs. Move any skills found in
.claude/skills/or.cursor/skills/into.agents/skills/. No format translation needed.Agents: Agent definitions are identical across IDEs. Consolidate into
.agents/agents/.Ignore patterns: If
.cursorignoreexists, extract the patterns into.agents/ignore. If.claude/settings.jsonhaspermissions.denywithRead()patterns, extract those too. Merge and deduplicate.Top-level instructions: Content from
CLAUDE.mdor.cursorrulesthat represents always-applied project instructions should become rules in.agents/rules/(withoutpaths:frontmatter so they apply always).MCP servers: Note any MCP configurations found but do not consolidate them yet (future rosetta feature).
Present your proposal to the user clearly:
I found the following agent infrastructure:
From .cursor/rules/:
- react-patterns.mdc (path-scoped: src/**/*.tsx) — React component conventions
- testing.mdc (always-apply) — Testing standards
From CLAUDE.md:
- Project instructions covering TypeScript, testing, and deployment
From .cursorignore:
- 5 ignore patterns
Proposed .agents/ structure:
.agents/rules/react-patterns.md ← from react-patterns.mdc (paths: instead of globs:)
.agents/rules/testing.md ← from testing.mdc
.agents/rules/project.md ← extracted from CLAUDE.md
.agents/ignore ← from .cursorignore
Shall I proceed?
Step 3: Create the .agents/ directory
After user approval:
- Create
.agents/with subdirectories:rules/,skills/,agents/ - Create
.agents/config.sh:ROSETTA_IDES=("claude" "cursor") ROSETTA_ARTIFACT_TYPES=("rules" "skills" "agents") - Write each proposed file, performing format translation as needed:
- Convert
globs:→paths:in rules from Cursor - Convert
.mdc→.mdextension - Add
alwaysApply: truenote as a comment for rules that had it - Preserve all content and comments
- Convert
- If ignore patterns were found, create
.agents/ignore
Step 4: Suggest next steps
After creating .agents/, tell the user:
- Review: Look through
.agents/to verify the consolidated content is correct - Sync: Run
/rosetta-syncto push the canonical versions back to all IDE directories - Commit: Add
.agents/to version control so the whole team benefits - Existing files: The original IDE-specific files are still in place — after verifying the sync works, you can optionally remove duplicates or let rosetta manage them going forward
Important notes
- Never delete existing files without explicit user approval
- Preserve content faithfully — adapt format (frontmatter keys, extensions) but don't alter the actual instructions/content
- Handle edge cases gracefully — if a file has unusual structure or you're unsure how to translate it, tell the user and ask
- Load
references/format-mapping.mdif you need the exact translation rules between IDE formats
Source: tannishmango/rosetta — distributed by TomeVault.