chemaclass/agnostic-ai/import

import

Captures what one tool already knows and makes it drive all of them, without retyping the conventions.

Steps

1. Confirm the CLI is present: agnostic-ai --version. If missing, run the install skill first.
2. Inventory what exists. Check for CLAUDE.md, AGENTS.md, GEMINI.md, CONVENTIONS.md, .cursor/rules/, .clinerules/, .devin/rules/, .continue/rules/, .github/copilot-instructions.md, and any .claude/{agents,skills,settings.json}.
3. Scaffold the config if absent: agnostic-ai init.
4. Import each source you found, one command per tool:

``bash agnostic-ai import claude # CLAUDE.md + .claude/{agents,skills,settings.json} agnostic-ai import codex # AGENTS.md, root and nested agnostic-ai import cursor # .cursor/rules/*.mdc agnostic-ai import cline # .clinerules/, .cline/agents/ agnostic-ai import windsurf # .devin/rules/ agnostic-ai import continue # .continue/rules/ ``

import writes spec files only. It never edits targets: or the rest of the config.

1. Commit the import before syncing, then sync as its own commit:

``bash git add .agnostic-ai/ agnostic-ai.yaml git commit -m "chore(agnostic-ai): import existing config" agnostic-ai sync git add -A git commit -m "chore(agnostic-ai): regenerate per-target configs" ``

Splitting them keeps the review honest: the first commit carries the real content, the second is mechanical (headers, key order) and can be skimmed.

1. Read the imported specs with the user. An imported rule often mixes several concerns because it came from one long file; split it into one file per rule so scoping and per-target emission work properly.

Notes

• Importing from several tools: run each import and its commit before the final sync.
• Round-trip is byte-stable: import strips the generated markers it wrote, so a later sync produces the same bytes.
• What import does not capture: helper scripts referenced from settings.json, statusline.sh, and ad-hoc config files. Keep those in git next to .agnostic-ai/. Files bundled inside a skill directory do round-trip.