rulesync-for-ai-codev

name	rulesync-for-ai-codev
description	Managing AI coding assistant rules, skills, and subagents with Rulesync

Rulesync

This project uses Rulesync to maintain a single source of truth for AI coding assistant configuration across tools (Claude Code, Copilot, Codex CLI, Cursor, and so on).

Structure

• .rulesync/rules/*.md: always-on context (loaded every session)
• .rulesync/skills/*/SKILL.md: on-demand context (invoked when needed)
• .rulesync/subagents/*.md: specialized assistant personas
• rulesync.jsonc: shared config; declares targets: ["copilot"] and the enabled features
• rulesync.local.jsonc: personal tool targets (gitignored); pick whatever you use locally
• .rulesync-version: pinned rulesync version

What gets committed

Copilot is the project's one shared target. Its generated output lives under .github/ (copilot-instructions.md, instructions/, agents/, skills/) and is checked in so Copilot Code Review has access to the project's rules and subagents on every PR. CI runs rulesync generate --check against this output to catch drift.

Generated output for any other target (.claude/, .cursor/, CLAUDE.md, .codex/, etc.) is gitignored. Never edit any generated output directly; rerun the generator instead.

The markdown linters (cspell, markdownlint-cli2, prettier) scan the whole repo, but each is configured to skip rulesync's generated outputs (.github/copilot-instructions.md and the .github/{instructions,agents,skills}/ trees). The sources under .rulesync/ are still linted along with everything else; only the generated copies are excluded. Without that exclusion the linters and rulesync would fight: prettier would reformat a generated file and the next rulesync generate would overwrite it, producing perpetual drift.

Making changes

When you edit anything under .rulesync/:

1. Lint first. .rulesync/**/*.md files go through markdownlint-cli2 and prettier like any other markdown. Run ./scripts/lint-files.sh --fix. Prettier may reformat them.
2. Then regenerate. ./scripts/rulesync-upgen.sh always refreshes the committed copilot output regardless of your personal rulesync.local.jsonc targets.
3. Commit both source and generated files.

Doing step 2 before step 1 produces drift between the source (now reformatted by prettier) and the generated output (still matches the pre-reformat source). CI will fail in that case.

To verify the committed copilot output is up to date without regenerating, run ./scripts/rulesync-upgen.sh --check.

Reference

See CONTRIBUTING.md for setup instructions and the Rulesync documentation for full details on rules, skills, subagents, targets, sources, and other features.