How every platform, tool, and data source connects to the EQ5 AI agent — what flows in, what flows out, and how it all stays in sync.
CLAUDE.md, team files, and any files you mention — before responding.CLAUDE.md, Claude calls MCP servers (Pipedrive, Slack, Notion, etc.) or REST APIs to pull live data. All tool calls are visible and require approval unless pre-allowed in settings.json../deploy.sh when ready../deploy.sh pushes updates to GitHub, which triggers Cloudflare Pages to publish the hub. Cloudflare Access enforces who can see each page — no separate auth code needed..claude/launch.json. Tool calls require approval unless pre-allowed in settings.json.eq5-sales repo stores everything: agent instructions, hub HTML, team profiles, Pipedrive exports, and weekly reports. Git is how changes get tracked and deployed.git via Bash — reads status, diffs, and log. You commit and push; Claude stages but does not push without confirmation./pipedrive skill — no exported files needed for current data./pipedrive skill, which loads curl examples and custom field mappings. Prefer live API over snapshot exports — snapshots go stale. Read-only in current setup.slack-feed.json for the homepage news ticker — separate from the live MCP connection.internal/campaign_processes/ for templates.zkasJapbZWbZ88TXj5oHSk) and Brandbook (MoIYxNnhadZgmS7uFXr4Bf). Always run /figma-use before calling use_figma../deploy.sh.internal/access_management/README.md for onboarding/offboarding runbooks. Per-client access is mirrored in hub/clients/{client}/access.md../deploy.sh.allow so Claude doesn't prompt you every time. Never auto-allow destructive operations (delete, send, post). Run /fewer-permission-prompts to get suggestions..sys-card block in training/index.html, update the name, category, connection type badge, and data chips to keep the team's mental model current.