Harness Init

Initialize codeharness in the current project. The CLI does the heavy lifting — your job is to run it and handle BMAD patches.

Step 1: Run the CLI

Run codeharness init and let it handle:

• Stack detection
• Dependency installation (with smart fallbacks — do NOT install deps manually)
• Docker/observability stack check
• State file creation
• Documentation scaffold
• Workflow template generation

codeharness init --json

Parse the JSON output to understand what happened. The output includes status, stack, stacks, dependencies, docker, otlp, documentation, and workflow fields.

If the state file already exists (re-init), the CLI handles preservation of enforcement config and verification_log automatically.

What if the user wants specific options?

• --no-observability — skip OTLP instrumentation
• --observability-backend none — disable observability entirely
• --otel-endpoint <url> — use remote OTLP endpoint
• --force — overwrite existing workflow file
• --frontend --no-database --api — set enforcement flags

Pass user preferences as CLI flags.

Step 2: BMAD Installation & Patches

The CLI does NOT handle BMAD installation or patch application. This is your responsibility.

Check for existing BMAD

1. If _bmad/ directory exists → preserve it, apply patches
2. If NO _bmad/ → run npx bmad-method install --yes --tools claude-code

Apply Harness Patches

Read each patch template from the codeharness plugin's templates/bmad-patches/ directory. Check if the patch marker already exists in the target file before applying.

Each patch is wrapped with markers:

<!-- CODEHARNESS-PATCH-START:{patch_name} -->
{patch content}
<!-- CODEHARNESS-PATCH-END:{patch_name} -->

If markers already exist, skip that patch. Patches are idempotent.

Step 3: Report

Read the version from codeharness --version for the report header. Output:

Harness Init — codeharness v{version}

[OK] Stack detected: {stack} ({indicator})
[OK] Docker: {running|missing}
[OK] Dependencies: {summary from CLI output}
[OK] BMAD: {installed|existing|standalone}
[OK] Documentation scaffold: {created|preserved}
[OK] Enforcement: frontend:{ON|OFF} database:{ON|OFF} api:{ON|OFF} observability:ON
[OK] Config: .claude/codeharness.local.md

→ Run /codeharness:harness-run to start autonomous execution.

Critical Rules

1. Do NOT run manual dependency checks via bash — codeharness init handles all deps
2. Do NOT edit the state file manually — the CLI manages .claude/codeharness.local.md
3. Do NOT check Docker, Showboat, Semgrep, BATS individually — the CLI checks them all in batch
4. Do NOT install tools into the project's .venv — the CLI uses system-level installers
5. Do NOT generate compose files — the CLI manages the shared observability stack
6. Do NOT hardcode version numbers — read from codeharness --version