benjaminshoemaker/configure-verification

Auto-detect verification commands and write .claude/verification-config.json. Runs silently with no user prompts — best-effort detection only.

Workflow

Copy this checklist and track progress:

Configure Verification Progress:
- [ ] Detect project directory and context
- [ ] Directory guard
- [ ] Detect package manager
- [ ] Auto-detect commands from project sources
- [ ] Conditional dev server detection
- [ ] Ensure .gitignore protection
- [ ] Write verification-config.json
- [ ] Output one-line summary

Project Directory

Use the current working directory by default.

If $1 is provided, treat $1 as the working directory and read files under $1 instead.

Context Detection

Determine working context:

1. If the working directory matches */features/*:
   • PROJECT_ROOT = parent of parent of working directory
2. If the working directory matches */plans/greenfield*:
   • PROJECT_ROOT = parent of parent of working directory
3. Otherwise:
   • PROJECT_ROOT = working directory

Directory Guard

Confirm PROJECT_ROOT exists and is writable. If not, stop and report the error.

Detect Package Manager

Determine package manager from lockfile presence (check in this order):

| Lockfile | Package Manager | |----------|----------------| | pnpm-lock.yaml | pnpm | | yarn.lock | yarn | | package-lock.json | npm |

If none found, default to npm if package.json exists, otherwise no package manager.

Auto-Detect Commands

Scan project sources in priority order. First source wins per command type. Parse files as structured data (JSON for package.json, not grep).

Priority Order

1. package.json scripts — Parse JSON, check for these script names:

   • test → {pm} test or {pm} run test
   • lint → {pm} run lint
   • typecheck or type-check or tsc → {pm} run typecheck (use actual script name)
   • build → {pm} run build
   • coverage or test:coverage → {pm} run coverage (use actual script name)
   • mutation or test:mutation or stryker → {pm} run {actual-script-name}

2. Makefile — Look for targets: test, lint, typecheck/check, build, coverage

   • Command format: make {target}

3. Taskfile.yml — Look for tasks with matching names

   • Command format: task {name}

4. justfile — Look for recipes with matching names

   • Command format: just {name}

5. Fallback: README.md / CONTRIBUTING.md — Scan for code blocks containing test/lint/build commands. Only use if no structured source found a command.

Important: Only record commands that were actually found. Do not guess or synthesize commands. If a command type is not found in any source, omit it from the config entirely.

Conditional Dev Server Detection

Only run this section if the active execution plan contains BROWSER: criteria.

Resolve the active execution plan in this order:

1. EXECUTION_PLAN.md in the working directory
2. plans/greenfield/EXECUTION_PLAN.md in the working directory (e.g., BROWSER:DOM, BROWSER:VISUAL, etc.).

If browser criteria exist, look for dev server scripts in package.json:

• dev → {pm} run dev
• start → {pm} run start
• serve → {pm} run serve

If found, include in config:

{
  "devServer": {
    "command": "{pm} run dev",
    "url": "http://localhost:3000",
    "startupSeconds": 10
  }
}

Use port 3000 as default. If the script content in package.json contains a different port (e.g., --port 5173), use that instead.

If no browser criteria in the active execution plan, omit devServer entirely.

Ensure .gitignore Protection

Check and protect verification-related files:

1. If .gitignore exists, ensure these entries are present:

   • .env.verification
   • .claude/verification/auth-state.json

2. If entries are missing, append them:

   # Verification credentials (never commit)
   .env.verification
   .claude/verification/auth-state.json
   

3. If .gitignore doesn't exist, skip (don't create one just for this).

Write Config

Write .claude/verification-config.json with only the keys where commands were found. Always include browser.tool: "auto".

Example — full detection:

{
  "commands": {
    "test": "npm test",
    "lint": "npm run lint",
    "typecheck": "npm run typecheck",
    "build": "npm run build",
    "coverage": "npm run coverage"
  },
  "browser": {
    "tool": "auto"
  }
}

Example — partial detection (no typecheck or coverage found):

{
  "commands": {
    "test": "npm test",
    "lint": "npm run lint",
    "build": "npm run build"
  },
  "browser": {
    "tool": "auto"
  }
}

Example — with dev server (browser criteria detected):

{
  "commands": {
    "test": "pnpm test",
    "lint": "pnpm run lint"
  },
  "devServer": {
    "command": "pnpm run dev",
    "url": "http://localhost:5173",
    "startupSeconds": 10
  },
  "browser": {
    "tool": "auto"
  }
}

Example — nothing detected:

{
  "commands": {},
  "browser": {
    "tool": "auto"
  }
}

Do NOT include auth, deployment, or devServer (unless browser criteria detected a dev server). These are deferred to /phase-checkpoint when needed.

Error Handling

| Situation | Action | |-----------|--------| | PROJECT_ROOT does not exist or is not writable | Stop immediately and report the path error to the user | | No package.json, Makefile, Taskfile.yml, or justfile found | Write config with empty commands object; report "no commands detected" | | package.json exists but contains invalid JSON | Report parse error with file path; skip package.json and try next source | | Lockfile detected but package manager binary is missing | Record the detected package manager but warn the user it is not installed | | .claude/ directory does not exist for writing config | Create .claude/ directory before writing verification-config.json |

Output

Print a single summary line:

Verification config: detected {list of found commands}. Missing: {list of not found}.

Examples:

Verification config: detected test, lint, build. Missing: typecheck, coverage.

Verification config: detected test, lint, typecheck, build, coverage, devServer. Missing: none.

Verification config: no commands detected.