docs(user-docs): validate every page against dev + sync changes since 09-30

[vybe]

Summary

Full /generate-user-docs run. Every page under docs/user-docs/ was re-checked against current dev code (routers, frontend views, MCP tools, agent server, deploy scripts and compose files), and behaviour shipped since the 09-30 sync (#3117) is now documented.

Scope: 99 pages edited, 1 new (guides/deploying/aws.md). FAQ grew from 523 to 616 questions; duplicates were merged and the index was regenerated from page headings.

Cross-cutting

• The positioning line ("the operating system for the AI-native company") and the three runtimes (Claude Code, Gemini CLI, OpenAI Codex) now appear on every page that defines Trinity.
• SQLite end-of-support is now past tense, and fresh hosted installs are described as running on a bundled PostgreSQL.
• New AWS spoke, plus a Vultr script install section. The README index now lists every deploy spoke.
• The MCP tool count is unified at 154; pages previously said 130, 142, 143 or 154.
• Removed docs for the /site/ agent-website proxy, which no longer exists in the code.

Newly documented

• Git pull sync and the "What is this repository?" picker
• Workspace Inbox, keyboard shortcuts and in-chat replies
• Asks: Discuss/Dismiss, "Something else", authoring limits. Platform alerts now behave as conditions.
• Cross-agent metric reads, idempotent human-facing sends, A2A x402 payments, permission withdrawal
• Event dispatch cap, chain depth through loops/schedules/events, the delegation receipt contract

Corrections

• /api/internal/* was documented as unauthenticated. It requires X-Internal-Secret.
• Agent delete is a soft delete (180 days), not a hard cleanup.
• Guardrails and read-only mode are Claude Code hooks and do not cover every runtime. Gemini cost is estimated.
• Public links are at /chat/{token}. WhatsApp proactive sends are not supported.
• docker system prune -f was called "safe"; it drops stopped agents from the roster.
• Log retention is 5 days, not 90.
• Four dead API paths fixed (/api/credentials/status, /api/monitoring/fleet-health, the OAuth callback, the 2FA route reference).

Verification

• Enterprise-docs guard regex: 0 hits.
• FAQ issue-number/codename grep: 0 real hits (only AES-256/SHA-256).
• Every relative link and anchor resolves.
• Deploy pages: private-IP, instance-path and AWS account/AMI ID greps all return 0 hits.
• The skill's tailnet grep matches generic Tailscale wording ("your tailnet") only. That wording was kept on purpose.
• The docs were not rendered.

Product issues surfaced (not fixed here)

• Gemini agents likely lose Workspace memory after the first turn: the backend treats only Codex as non-resuming, but the Gemini headless path ignores the resume id.
• vultr-vendor-data.sh installs releases/latest (v0.9.5), which rejects --cloud vultr.
• startup.sh calls /api/internal/decrypt-and-inject, which doesn't exist.
• A rotated .env value never reaches an already-rendered .mcp.json server.
• Stale UI/help copy:
  • the set_agent_github_pat description still says a restart is needed
  • "Clear All" says it permanently deletes, but it only hides
  • .mcp.json shows an Edit button that can't save
  • the CLI --repo help says "public" only

🤖 Generated with Claude Code

[trinity-ability]

merge-train: batch validated on train/20261007-0858 (train #3329, green)

[vybe]

merge-train (2026-10-07): merged last on train #3329. Pushed dd4eac5eb first, a dev merge in which four user-docs files (dynamic-dashboards.md, chat-api.md, faq/mcp-and-api.md, integrations/mcp-server.md) conflicted with #3294, #3297, #3299 and #3303, which had documented their own changes on the same lines. I resolved each hunk by hand, keeping both sides' facts once rather than keeping both blocks verbatim. Worth a skim.