greenfield（greenfield 搜索）

Greenfield Setup

Configures Claude Code for a new project. Everything is generated on the fly based on the actual stack and developer answers — not copied from generic templates.

Scope: Generates project foundation only:

• CLAUDE.md,.claude/rules/,.claude/hooks/, settings.json, settings.local.json,.gitignore
• Optional: git init + first commit, GitHub Actions CI workflows

Run /workflow after this to install development commands and agents.

Step 1 — Preflight & Stack Detection

Check whether any files that greenfield would generate already exist. Scan for:

• CLAUDE.md
• .claude/settings.json
• .claude/settings.local.json
• .claude/rules/ (any .md files inside)
• .claude/hooks/ (any .sh files inside)

Ignore the .claude/skills/ directory — that is where installed skills live and is not a conflict.

If no conflicting files are found, proceed silently. If any conflicting files exist, use AskUserQuestion to show the list and ask:

"The following files already exist and will be overwritten: - [list each conflicting file] Continue and overwrite these, or run /brownfield instead to analyze what's already here?" Choices: Continue / Abort

Wait for confirmation before proceeding. If they choose Abort, stop.

Run stack detection:

bash "${CLAUDE_SKILL_DIR}/scripts/detect-stack.sh"

The script outputs JSON to stdout with: stack, framework, lang, pkg_manager, formatter, test_runner, has_typescript. Parse and store all values — they will be used as environment variables for the generator scripts.

If stack is "unknown", ask the developer:

"I couldn't detect your stack. What are you building? (e.g. Next.js, React, Node.js API, Python, Go, Rust,.NET)"

Update the detection values based on their answer.

Step 2 — Check for Previous Configuration

Before starting the interview, check if this skill has been run before:

SKILL_NAME="greenfield" \
bash "${CLAUDE_SKILL_DIR}/scripts/load-config.sh"

If the output is a non-empty JSON object (not {}), previous answers exist. Present them:

Previous greenfield configuration found:

  Project:       [PROJECT_NAME] — [PROJECT_DESCRIPTION]
  Stack:         [STACK]
  Agent teams:   [AGENT_TEAMS]
  Commit style:  [COMMIT_STYLE]
  CI/CD:         [CI_CD]

Would you like to:
  a) Update — regenerate all artifacts using these settings
  b) Reconfigure — start the interview from scratch

Use AskUserQuestion and wait for the response.

• If Update: load all saved values into environment variables and skip to Step 3.
• If Reconfigure: proceed to the full interview below.

If no previous config exists, proceed directly to the interview.

Step 2b — Developer Interview

Load the question set from: ${CLAUDE_SKILL_DIR}/references/interview.md

Ask questions one at a time using AskUserQuestion. This provides a cleaner experience — the developer sees one focused question with clear choices instead of a wall of text.

For each question:

1. Use AskUserQuestion with the question text and suggested options/defaults
2. Wait for the response
3. Store the answer in the corresponding environment variable
4. Move to the next question

Ask all 7 questions (Q1–Q7) in order. Ask Q6b only if STACK == python — skip it for every other stack and leave TYPECHECKER empty. Skip Q7's follow-up if the developer answers "no" to CI/CD. Do not proceed to Step 2c until all questions are answered.

Step 2c — Save Configuration

After the interview (or after loading previous config for an update), save the current answers:

SKILL_NAME="greenfield" \
CONFIG_JSON='{"PROJECT_NAME":"[val]","PROJECT_DESCRIPTION":"[val]","STACK":"[val]","FRAMEWORK":"[val]","LANG":"[val]","PKG_MANAGER":"[val]","FORMATTER":"[val]","TEST_RUNNER":"[val]","TYPECHECKER":"[val or empty]","HAS_TYPESCRIPT":"[val]","GIT_SETUP":"[val]","AGENT_TEAMS":"[val]","COMMIT_STYLE":"[val]","CI_CD":"[val]","CI_WORKFLOWS":"[val]"}' \
bash "${CLAUDE_SKILL_DIR}/scripts/save-config.sh"

Replace [val] with actual values. This writes to .claude/skill-config.json so future runs can reuse these answers.

Step 3 — Load Stack Knowledge

Read the appropriate stack reference file based on the confirmed stack:

Extract the command mappings (INSTALL_CMD, DEV_CMD, BUILD_CMD, TEST_CMD, LINT_CMD) from the stack reference file based on the detected package manager.

Step 3a — Security Tools Check

Check whether security tools used by the quality gate are installed:

STACK="[confirmed stack]" \
bash "${CLAUDE_SKILL_DIR}/scripts/check-security-tools.sh"

The script outputs JSON with the status of gitleaks, semgrep, trivy, and oxlint (oxlint is skipped for non-Node stacks).

If all relevant tools are installed, print a brief confirmation and continue:

✅ Security tools: gitleaks ✓  semgrep ✓  trivy ✓

If any tools are missing, list only the missing ones with install instructions:

⚠️  Some security tools used by the quality gate are not installed.
The quality gate will skip these checks until they're installed.

Recommended tools:

  gitleaks — Secret/credential detection (finds API keys, tokens, passwords)
    https://github.com/gitleaks/gitleaks#installing
    brew install gitleaks

  semgrep — SAST scanner (injection, XSS, OWASP Top 10)
    https://semgrep.dev/docs/getting-started/
    pip install semgrep  /  brew install semgrep

  trivy — Vulnerability scanner (dependencies, containers, IaC)
    https://trivy.dev/latest/getting-started/installation/
    brew install trivy

  oxlint — Fast linter for JavaScript/TypeScript (Node stacks only)
    https://oxc.rs/docs/guide/usage/linter
    npm install -g oxlint

Install them anytime — the quality gate will pick them up automatically.

Continue to Step 5.

Step 5 — Generate All Artifacts

Run each script with the correct environment variables. All scripts require PROJECT_DIR. Scripts can run in parallel — they write to separate paths and have no interdependencies.

5a — Settings

PROJECT_DIR="$PWD" \
AGENT_TEAMS="[true/false from Q5]" \
PKG_MANAGER="[from detection]" \
bash "${CLAUDE_SKILL_DIR}/scripts/generate-settings.sh"

5b — Rules

PROJECT_DIR="$PWD" \
STACK="[confirmed stack from Q3]" \
bash "${CLAUDE_SKILL_DIR}/scripts/generate-rules.sh"

5c — Hooks

PROJECT_DIR="$PWD" \
STACK="[confirmed stack from Q3]" \
PKG_MANAGER="[from detection]" \
FORMATTER="[from detection]" \
TEST_RUNNER="[from detection, e.g. vitest/jest/pytest/go-test]" \
TYPECHECKER="[from Q6b, python only — otherwise leave empty]" \
bash "${CLAUDE_SKILL_DIR}/scripts/generate-hooks.sh"

5d — CLAUDE.md

PROJECT_DIR="$PWD" \
PROJECT_NAME="[from Q1]" \
PROJECT_DESCRIPTION="[from Q2]" \
STACK="[confirmed stack from Q3]" \
INSTALL_CMD="[from stack reference]" \
DEV_CMD="[from stack reference]" \
BUILD_CMD="[from stack reference]" \
TEST_CMD="[from stack reference]" \
LINT_CMD="[from stack reference]" \
COMMIT_STYLE="[from Q6]" \
bash "${CLAUDE_SKILL_DIR}/scripts/generate-claude-md.sh"

5e —.gitignore

PROJECT_DIR="$PWD" \
bash "${CLAUDE_SKILL_DIR}/scripts/generate-gitignore.sh"

Each script outputs JSON status to stdout (e.g. {"status":"ok","files":[...]}). Check that all return "status":"ok". If any fail, report the error and stop.

Step 6 — Git Setup (if requested)

Only if the developer chose init in Q4.

Use the commit message based on Q6:

• Conventional: chore(claude): initialize Claude Code configuration
• Free-form: Initialize project with Claude Code configuration

git init
git add CLAUDE.md .claude/ .gitignore
git commit -m "[commit message from above]"

Step 7 — CI/CD (if requested)

Only if the developer chose yes in Q7.

Load templates from: ${CLAUDE_SKILL_DIR}/references/ci-templates.md

Generate .github/workflows/ files for each selected workflow (lint, test, build, deploy). Replace placeholder values ($PKG_MANAGER, $INSTALL_CMD, $LINT_CMD, $TEST_CMD, $BUILD_CMD) with the actual values from stack detection and the stack reference.

If git was initialized in Step 6, stage and commit the CI files:

git add .github/
git commit -m "ci: add GitHub Actions workflows"

Step 8 — Completion

Tell the developer what was created:

✅ Greenfield Setup Complete — [Project Name]

Generated:
- CLAUDE.md
- .claude/settings.json        (agent teams: yes/no)
- .claude/settings.local.json
- .claude/rules/               [list files]
- .claude/hooks/               [list scripts]
- .gitignore                   (updated)

Next steps:
1. Run /workflow to install development commands and agents
2. Restart Claude Code to load the new configuration

If git was initialized, include the commit hash. If CI/CD was scaffolded, list the workflow files created.

Reference Files

• Full interview questions: ${CLAUDE_SKILL_DIR}/references/interview.md
• Artifact content specs: ${CLAUDE_SKILL_DIR}/references/artifact-specs.md
• CI workflow templates: ${CLAUDE_SKILL_DIR}/references/ci-templates.md
• Stack conventions: ${CLAUDE_SKILL_DIR}/references/stacks/