Update commands and docs for campaign snapshots

[thiagoc7]

Summary

• weekly-report: add steps 4d-4g (campaign snapshots, action proposals, dedup check, continuity with previous insights) + suggested_command in insights
• monthly-report: add mandatory body structure (matching weekly-report pattern) and steps 4d-4g
• analyze-performance: add snapshot and proposal steps
• v4-check: add campaign snapshots (no proposals — V4 executes)
• CLAUDE.md: add Diesel BI persistence section with health score rules, proposal rules, and dedup guidelines
• guides/integracao-diesel-bi.md: update from 5 to 13 tools, add CampaignSnapshot and ActionProposal models, update workflow and UI docs

Test plan

• Run /weekly-report and verify snapshots + proposals saved to Diesel BI
• Verify insights include suggested_command
• Verify dedup check works (run report 2x)
• Monthly report follows mandatory body structure

🤖 Generated with Claude Code

[kLOsk]

Hey @thiagoc7, thanks for the PR and for putting real time into using AdLoop — always good to see it used in production. I've gone through the full diff and want to give detailed feedback.

Why I can't merge this as-is
CLAUDE.md duplicates adloop.mdc
The 442-line CLAUDE.md is a near-complete copy of the existing .cursor/rules/adloop.mdc — safety rules, orchestration patterns, GAQL reference, character limits, marketing best practices. Maintaining the same content in two files means every future change needs to happen twice, and they'll drift apart. I'd need a solution where Claude Code rules derive from or reference the single source of truth rather than duplicating it.

Client-specific content in an open-source repo
Most slash commands are built around a specific client ("Disbra") and depend on an external "Diesel BI" MCP server that isn't part of AdLoop:

weekly-report.md, monthly-report.md, analyze-performance.md, v4-check.md — all hardcode company_slug "disbra" and Diesel BI persistence steps
historico.md — references [DISBRA] and [V4] campaign naming conventions
v4-check.md — evaluates a specific agency ("agência V4")
These commands won't function for anyone else installing AdLoop. They're great workflow automation for your setup, but they belong in your private project repo, not in the open-source tool.

Language consistency
7 of 9 command files are in Portuguese. The entire AdLoop codebase (README, docs, source code, docstrings, cursor rules) is in English. If we ship slash command templates, they need to be in English to match.

Python version downgrade (3.14 → 3.11)
The .python-version change isn't explained and would change the project baseline for all contributors.

Hardcoded local path
CLAUDE.md contains /Users/thiago/Projects/adloop which won't work for anyone else.

What I've already cherry-picked
Some pieces from this PR were genuinely useful, so I've applied them directly:

config.py post_init — the GA4 property ID auto-prefix (properties/) is a real QoL fix. Applied.
.gitignore additions — reports/, guides/, .playwright-mcp/. Applied.
CLI wizard Claude Code snippets — noted for our upcoming Claude Code plugin work (the idea of generating claude mcp add and .mcp.json snippets alongside the existing Cursor snippet is exactly right).
Slash command concept — the idea of shipping generic command templates (like your create-ad.md and diagnose-tracking.md) is a good one. I've added it to the Claude Code plugin roadmap. They'll be English-only, generic versions derived from the existing orchestration patterns.
Closing this PR
Since I've already incorporated the reusable parts and the rest is too client-specific to merge, I'm going to close this. No hard feelings — it's clear you've done real work here, and the Diesel BI + AdLoop combo you've built for your client is a solid use case. It just doesn't fit as a contribution to the open-source project itself.

If you want to contribute to the Claude Code plugin when we build it properly (single source of truth for rules, generic English commands, no external dependencies), I'd welcome that.