Your team pays for Claude Code.
I make it part of how they ship.

The rules · CLAUDE.md (excerpt)

## Prohibited actions
- If something that should work with real data isn't working,
  fix it with real data. No mock, stubbed or simulated data
  unless explicitly requested.
- Never write TODO tasks in the code itself.
- Validate that all tests pass before marking a task complete.
- Never mark a task complete without the user's approval.

## Modes
EXPLORE → PLAN → IMPLEMENT.
No files change in EXPLORE. Plans need explicit approval.
Silence is not approval.

The guardrails · ~/.claude/settings.json (excerpt)

"hooks": {
  "UserPromptSubmit": [{ "hooks": [{
    "type": "command",
    "command": "python3 ~/.claude/scripts/wake_up.py --file",
    "timeout": 5000 }] }],
  "Stop": [{ "hooks": [{
    "type": "command",
    "command": "python3 ~/.claude/scripts/session_close.py",
    "timeout": 5000 }] }],
  "PreToolUse": [{ "matcher": "*", "hooks": [{
    "type": "command",
    "command": "node monitor/hook-handler.js PreToolUse" }] }]
}

The integration · ad-operations MCP server (excerpt)

@mcp.tool
@tool_with_telemetry(category="campaign-management")
async def get_campaign(campaign_id: str, ctx: Context = None) -> dict:
    """Campaign ID → full config: paths, cost model,
    traffic source, country, URL.

    Use when: pre-launch verification → confirm path
      structure and offer assignment.
    Use when: diagnosing path issues → inspect default
      and conditional paths.
    Use when: updating a campaign → fetch the full entity
      first (the API requires all fields on PUT).
    """
    return await tracker.get(f"/campaign/{campaign_id}")

@mcp.tool
@tool_with_telemetry(category="campaign-management", is_write=True)
async def create_campaign(...): ...