Hooks Management and Analytics gets an AI Control tab

Hooks — every agent's automations in one dialog — open it from Tools → Hooks in the AI group or Settings → Agents → Open hook manager, and every hook your agents have configured is listed in one place: Claude, Codex, Gemini, Kimi, Antigravity, Cursor, Kiro, Devin, Qwen, Qoder, Grok and more, grouped by agent with the config profile each group reads. Each row says when it runs (before a tool runs, after it finishes, when a session starts…), which config file it came from, whether it's user or project scope and who owns it — and you can search or filter by name, agent, scope or status, including a view of just the entries that need attention.

Turn a hook off and back on with a switch — the dialog edits the agent's real config file, offers Undo after saving, and names the provider's next step when a re-enabled hook needs trust or a reload first. Pause my hooks captures the currently enabled set and disables it in one switch; unpausing restores exactly that set and leaves hooks that were already off alone.

Nothing you're not allowed to touch is touched — 1DevTool's own session-tracking integrations, organization policies and plugin-owned hooks show read-only with the reason and are never rewritten. Every change is journaled in the Changes tab, so a bad edit is recoverable, and any hook's config file opens in the editor or reveals on disk straight from its row.

Analytics gets an AI Control tab — pick a session and read how it was actually spent: its tool histogram, the MCP servers it called, the skills it fired, how long it ran and where the tokens went, with the bill split into input, output, cache and reasoning plus an estimate of tool-turns vs plain text-turns.

Orchestrate view grades your runs' health — team, swarm and headless runs in range with pass rate, timeouts and interruptions, median and p90 duration, tokens burned, and plain-language observations instead of a score. Compare view puts two agents side by side on the same range and the same table.

The scan is lazy and honest — it only runs while the tab is open, reads a bounded slice of recent session files, and tells you when coverage was partial instead of silently dropping data.

Add and edit MCP servers from Settings — Settings → MCP → Add server installs a server into every supported client from one form, writing the right native shape for Claude, Codex, Gemini, Kimi, OpenCode, Grok, Hermes, Cursor, Oh My Pi, Kiro and Devin. Local or remote: a command with args and environment, or a URL with headers, with the enabled flag honored where the client supports it.

Edit a server in place — any third-party entry opens back up in the same form with its real values decoded, no config-file spelunking; servers the 1DevTool installer owns are listed and marked rather than hidden, and disabled entries carry a label instead of pretending to be live.

Diagnosis checks default to the cheapest model — an Autopilot checker left on a blank model now runs on Claude's Haiku or Codex's gpt-5.6-luna instead of the account's priciest default (measured ~50× cheaper per check), and the model field names what will actually run.

AI Diagnosis reads as a panel, not a wall — the tab is split into Overview (just the switches), Explain (what each concept means, one click from a “Why?”) and History (Autopilot's decisions on this machine, newest first).

The top bar makes room for Analytics — Prompts and Memories now share a Context dropdown and Skills, MCP and Hooks share a Tools dropdown, so the Analytics button is pinned on the bar by default; every entry is still toggled individually in Settings → Layout.