autostackrc

Detect your tech stack → generate tailored rules & commands for your AI code assistants.

autostackrc generates rules and slash-commands specific to your stack and writes them in the native format of each assistant — Claude Code, Cursor, GitHub Copilot, and opencode.

📖 Docs & portal: https://cesarzuniga.github.io/autostackrc (deployed via GitHub Pages from docs/).

npx autostackrc

How it works

1. Detect — scans package.json, config files, and file extensions to identify your stack.
2. Assemble — builds a baseline of rules + commands from static templates (offline, auditable).
3. Enrich — hands the baseline to your installed AI coder CLI (using its existing login session, no API key) to make the rules specific and correct for your stack.
4. Write — projects one agent-agnostic result into every selected assistant's files.

If no coder CLI is found (or --no-llm), it emits the static templates as-is — never hard-fails.

Output per assistant

Assistant	Rules	Commands
Claude Code	CLAUDE.md (managed block)	.claude/commands/*.md
Cursor	.cursor/rules/*.mdc (one per tech, globs/alwaysApply)	.cursor/commands/*.md
GitHub Copilot	.github/copilot-instructions.md + .github/instructions/*.md	.github/prompts/*.prompt.md
opencode	AGENTS.md (managed block)	.opencode/command/*.md

Shared files (CLAUDE.md, AGENTS.md, copilot-instructions.md) are edited inside an <!-- autostackrc:start -->…<!-- autostackrc:end --> block, so re-runs update only that block and leave your own content untouched. A autostackrc-lock.json records every file written.

Enrichment without an API key

autostackrc does not call any API directly. It detects an installed coder CLI in this order and runs it headless, feeding the prompt on stdin and parsing the JSON it returns:

• claude (Claude Code) — claude -p
• opencode — opencode run
• cursor-agent (Cursor) — cursor-agent -p

Whatever session that CLI is already logged into is what does the work. Force one with --coder.

Options

npx autostackrc                        Detect, generate, write (interactive)
  -y, --yes         Skip confirmation prompts
  --dry-run         Show plan, write nothing
  -a, --agent       Target subset: claude-code cursor copilot opencode
  --no-llm          Skip enrichment, emit static templates only
  --coder <id>      Force coder CLI: claude, opencode, cursor-agent
  --model <id>      Model hint passed to the coder CLI (if supported)
  -v, --verbose     Show traces and error detail
  -h, --help        Show this help

Requirements

Node.js >= 22.6.0 (uses native TypeScript type-stripping).

Extending

• Add a technology: create templates/rules/<id>.md (frontmatter + rule body) and add a seed (detect rules + command ids) to TECH_SEEDS in catalog.ts.
• Add a command: create templates/commands/<id>.md (frontmatter + prompt body); mark it generic: true to emit for every stack, or wire its id into a tech's commands in catalog.ts.
• Add an assistant: add a writer in writers/ and register it in writer.ts + agents.ts.