onboarding: add Herdr as a supported multiplexer

onboarding/MAINTAINING.md

@@ -14,7 +14,7 @@ not by copying secrets or by being granted access to the PI's machines.
 ## Architecture (the mental model the kit assumes)
 
 - **One real environment, on Linux** (the member's office PC). The agent stack — Claude
-  Code, agent-skills, superpowers, tmux — lives here and only here.
+  Code, agent-skills, superpowers, and a multiplexer (tmux or Herdr) — lives here and only here.
 - **Laptops are dual-role ("mixed by purpose"):** thin client (ssh/mosh + `tmux attach`)
   into the Linux box for heavy work, *plus* native Obsidian for notes. A laptop needs only
   tailscale + a terminal + Obsidian.
@@ -33,6 +33,7 @@ not by copying secrets or by being granted access to the PI's machines.
 | `onboarding/vault-structure.md` | DRGScripts | Vault tree + scaffold + which seeds go where. | The vault layout changes. |
 | `onboarding/vault-seeds/**` | DRGScripts | Real files copied into the member's vault (operating manual, wiki index, cheatsheets, tmux.conf). | The seeds drift from the PI's (sanitized) originals. |
 | `onboarding/dotfiles/tmux.conf` | DRGScripts | The group tmux config. | The PI's `~/.tmux.conf` changes. |
+| `onboarding/dotfiles/herdr-config.toml` | DRGScripts | The group Herdr config (for members who use Herdr instead of/besides tmux). | The PI's `~/.config/herdr/config.toml` changes. |
 | `ARC/CLAUDE.md` | DRGScripts | Existing ARC project guide — referenced, not duplicated. | (Maintained independently.) |
 
 ## Decisions & rationale
@@ -47,7 +48,9 @@ not by copying secrets or by being granted access to the PI's machines.
 | Statusline | the **simple context-% script**, not the PI's `~/agents` auto-handoff variant | Self-contained; no extra infra to stand up. |
 | Seeds | **sanitized**, real files under `vault-seeds/` | Easy `cp` into place; the PI's real remote-dev note (tailnet IPs, VPN endpoint, cluster/exit-node config) is excluded — a generic `Remote Dev — Pattern` replaces it. |
 | First pass | **Claude Code only** | Smaller surface; the deferred list below grows it later. |
-| Silent-stall detection | **cc-watchdog** (`onboarding/watchdog/`): timer-driven dead-man's switch, not hook-driven heuristics | CC-hook watchers only wake at turn ends, so they share the session's blind spot (a 36 h TA die-out in 2026-07 motivated this). Deadlines declared by the session itself make false positives ~zero (a 2.75 h quiet suite run is legitimate work); a 6 h notify-only backstop covers sessions that never declared. Unlike the PI-local auto-handoff/Phoenix stack, this ships to members: standalone (tmux + coreutils + systemd user timer + one Slack webhook). |
+| Multiplexer | **Herdr recommended; tmux also supported** (ONBOARDING step 7) | Herdr is agent-state-aware (mouse-first sidebar showing each agent's live state), which the cc-watchdog guard and the agent-skills auto-handoff/Phoenix watchers gate on natively — so it's the recommended default. tmux stays fully supported for members who deliberately prefer a keyboard-first multiplexer. |
+| Herdr `settings.json` hook | **member runs `herdr integration install claude`** | Unlike the PI's literal settings.json, the herdr agent-state hook is generated by the installer into the member's own `~/.claude/settings.json` — no personal infra to copy. |
+| Silent-stall detection | **cc-watchdog** (`onboarding/watchdog/`): timer-driven dead-man's switch, not hook-driven heuristics | CC-hook watchers only wake at turn ends, so they share the session's blind spot (a 36 h TA die-out in 2026-07 motivated this). Deadlines declared by the session itself make false positives ~zero (a 2.75 h quiet suite run is legitimate work); a 6 h notify-only backstop covers sessions that never declared. Unlike the PI-local auto-handoff/Phoenix stack, this ships to members: standalone (tmux/Herdr + coreutils + systemd user timer + one Slack webhook). |
 
 ## Deferred — "later" (and how to un-defer)
 

onboarding/ONBOARDING.md

@@ -10,9 +10,9 @@
 ## What you're building
 
 - **One real environment, on your office Linux PC** — the agent stack (Claude Code,
-  agent-skills, superpowers, tmux) lives here and only here.
+  agent-skills, superpowers, Herdr) lives here and only here.
 - **Your laptop (Win/Mac) is a thin client + Obsidian** — you SSH/mosh into the Linux PC
-  and `tmux attach`, and you run Obsidian locally on the Dropbox-synced vault.
+  and attach your Herdr session (or tmux), and you run Obsidian locally on the Dropbox-synced vault.
 - This is **your own independent setup**: your own Tailscale tailnet, your own Obsidian
   vault, your own Claude account. Nothing here grants access to anyone else's machines.
 
@@ -77,12 +77,43 @@ folders there; those are **git-ignored** — don't commit them. Update gstack la
 "use `/browse` for all web browsing" rule) comes from the CLAUDE.global.md merge in step 8, so
 you don't add it by hand here.
 
-### 7. tmux config
+### 7. Terminal multiplexer — Herdr (tmux also supported)
+A multiplexer keeps your panes (and long agent sessions) alive across
+disconnects, and is what the cc-watchdog guard (step 13) drives.
+**Use [Herdr](https://herdr.dev)** — it's the group's recommended default because
+it's agent-aware (shows each agent's live state in a sidebar) and the cc-watchdog
+guard and the PI's watchers gate on that state natively. tmux is also fully
+supported if you deliberately prefer a keyboard-first, ubiquitous multiplexer.
+
+**Herdr** (recommended):
+```bash
+curl -fsSL https://herdr.dev/install.sh | sh     # installs the `herdr` binary
+herdr integration install claude                 # wires the agent-state hook into ~/.claude/settings.json
+mkdir -p ~/.config/herdr                          # ensure the config dir exists before copying
+cp <path-to-this-DRGScripts-clone>/onboarding/dotfiles/herdr-config.toml ~/.config/herdr/config.toml
+```
+The `/herdr` control skill ships in the agent-skills clone from step 5, so
+Claude Code can drive panes/tabs directly. Launch with `herdr`, then start
+`claude` inside a pane; detach with `prefix+q` (prefix is `ctrl+b`), `prefix+?`
+lists all bindings.
+
+**tmux** (alternative — if you deliberately prefer it):
 ```bash
 cp <path-to-this-DRGScripts-clone>/onboarding/dotfiles/tmux.conf ~/.tmux.conf
 ```
 Prefix is `C-a`. (Optional persistence plugins need `tpm` — see the comments in the file.)
 
+The cc-watchdog stall-guard (step 13) works under both Herdr and tmux. (The PI's optional
+auto-handoff / Phoenix watchers — deferred for members, see MAINTAINING — are
+likewise multiplexer-aware if you enable them later.)
+
+Optional remote-attach aliases (add to `~/.bashrc`), to connect and attach on a
+workstation named `ol`:
+```bash
+alias olh='herdr --remote ol'                       # Herdr (needs herdr installed on ol)
+alias olt='ssh -t ol "tmux attach || tmux new"'   # tmux
+```
+
 ### 8. Global CLAUDE.md
 Merge [CLAUDE.global.md](./CLAUDE.global.md) into your `~/.claude/CLAUDE.md` (have Claude
 Code merge it, preserving any lines you've already added). Fix the **Obsidian Vault path**
@@ -218,7 +249,12 @@ rules are the prevention. Details: `onboarding/watchdog/README.md`.
    ```
    (macOS: skip — use the native Terminal; `brew install mosh` if you want mosh.)
 2. **Tailscale** — install the GUI app, sign in to the same tailnet, confirm *Connected*.
-3. **Terminal into the Linux PC** (run agents on the remote, never the laptop):
+3. **Terminal into the Linux PC** (run agents on the remote, never the laptop). With Herdr
+   (recommended) one command attaches your remote session — start `claude` inside a pane:
+   ```bash
+   herdr --remote <office-pc>      # your user + tailnet host name; needs herdr on both ends
+   ```
+   Prefer tmux? Attach the classic way instead:
    ```bash
    ssh <you>@<office-pc>      # your user + tailnet host name
    tmux attach -t cc || tmux new -s cc
@@ -236,7 +272,8 @@ rules are the prevention. Details: `onboarding/watchdog/README.md`.
 ## C. Verification smoke test
 
 - [ ] `tailscale status` shows your tailnet and this host.
-- [ ] From the laptop: `ssh`/`mosh` into the Linux PC, `tmux attach` works.
+- [ ] From the laptop: `ssh`/`mosh` into the Linux PC and attach your session
+      (`herdr --remote`, or `tmux attach`) works.
 - [ ] A Claude Code session lists the gstack skills from step 6 (type `/` and look for
       `/browse`, `/review`, `/ship`, …); superpowers loads.
 - [ ] The status line shows the model name + context-window %.

onboarding/dotfiles/herdr-config.toml

@@ -0,0 +1,26 @@
+# Herdr config — sensible defaults for autonomous multi-agent work.
+# Place at ~/.config/herdr/config.toml. Herdr runs fine without it; these are
+# only deltas from the built-in defaults. See every knob: `herdr --default-config`.
+# Reload into a running server after edits: `herdr server reload-config`.
+
+[ui]
+# Float blocked / needs-attention agents to the top of the sidebar instead of
+# grouping by space. Handy when several agents run at once.
+agent_panel_sort = "priority"
+
+[ui.toast]
+# Notify when a background agent changes state (done / blocked). "terminal"
+# routes the notification to the terminal you're currently attached to, so it
+# reaches you even when you attach remotely over SSH (unlike "system", which
+# fires on the host's own desktop).
+delivery = "terminal"
+
+[experimental]
+# Preserve pane screen contents across a full server restart, so a long run's
+# output isn't lost if the server bounces.
+pane_history = true
+
+[advanced]
+# Long autonomous sessions emit a lot of output — keep more scrollback for
+# review (50MB, up from the 10MB default).
+scrollback_limit_bytes = 50000000