---

description: Detect a project's stack and produce a dry-run ECC onboarding plan using the repository's install manifests and stack mappings.

/project-init

Create a safe, reviewable ECC onboarding plan for the current project. This command should start in dry-run mode and only write files after explicit user approval.

Usage

/project-init
/project-init --dry-run
/project-init --target claude
/project-init --target cursor
/project-init --skills continuous-learning-v2,security-review
/project-init --config ecc-install.json

Safety Rules

1. Default to dry-run. Do not modify CLAUDE.md, settings files, rules, skills, or install state until the user approves the concrete plan.
2. Preserve existing project guidance. If CLAUDE.md, .claude/settings.local.json, .cursor/, .codex/, .gemini/, .opencode/, .codebuddy/, .joycode/, or .qwen/ already exists, inspect it and propose a merge/append plan instead of overwriting.
3. Use ECC's installer and manifest tooling. Do not hand-copy files or clone arbitrary remotes as an install shortcut.
4. Keep permissions narrow. Any generated settings should match detected build/test/lint tools and avoid broad shell access.
5. Report exactly what would change before applying anything.

Detection Inputs

Read the current project root and detect stack signals from:

• package manager files: package.json, package-lock.json, pnpm-lock.yaml, yarn.lock, bun.lockb
• language manifests: pyproject.toml, requirements.txt, go.mod, Cargo.toml, pom.xml, build.gradle, build.gradle.kts
• framework files: next.config.*, vite.config.*, tailwind.config.*, Dockerfile, docker-compose.yml
• ECC config: ecc-install.json
• optional stack map: config/project-stack-mappings.json in the ECC repo

When the ECC checkout is available, use config/project-stack-mappings.json as the stack-to-rules/skills reference. If the file is unavailable, fall back to the installed ECC manifests and explicit user choices.

Planning Flow

1. Identify the target harness. Default to claude unless the user asks for cursor, codex, gemini, opencode, codebuddy, joycode, or qwen.
2. Detect stacks from project files and show the evidence for each match.
3. Resolve the smallest useful ECC plan:
   • project has an ecc-install.json: node scripts/install-plan.js --config ecc-install.json --json
   • user named a profile: node scripts/install-plan.js --profile <profile> --target <target> --json
   • user named skills: node scripts/install-plan.js --skills <skill-ids> --target <target> --json
   • only language stacks are detected: use the legacy language install dry-run with those language names
4. Run a dry-run apply command before writing:

node scripts/install-apply.js --target <target> --dry-run --json <language-or-profile-args>

5. Summarize detected stacks, selected modules/components/skills, target paths, skipped unsupported modules, and files that would be changed.
6. Ask for approval before applying the non-dry-run command.

Output Contract

Return:

1. detected stack evidence
2. proposed target harness
3. exact dry-run command used
4. exact apply command to run after approval
5. files/directories that would be created or changed
6. warnings about existing files, broad permissions, missing scripts, or unsupported targets

CLAUDE.md Guidance

If the user wants a CLAUDE.md starter, generate it separately from the installer plan and keep it minimal:

• build command, if detected
• test command, if detected
• lint/typecheck command, if detected
• dev server command, if detected
• repo-specific notes from existing package scripts or manifests

Never replace an existing CLAUDE.md without showing a diff and receiving approval.

Related

• config/project-stack-mappings.json for stack-to-surface hints
• scripts/install-plan.js for deterministic plan resolution
• scripts/install-apply.js for dry-run and apply operations
• /ecc-guide for interactive feature discovery before installing