collab-stack: import collab stack with local repairs

Start from the pinned `origin/dev/friel/collab-stack` snapshot and fold in the
local follow-up repairs that made that imported stack usable here: watchdog
spawn/registration plumbing, deferred-tool fallback behavior, collab discovery
fixture alignment, schema mirroring, and subagent-panel fixes.

Original imported source:
- source ref: `refs/remotes/origin/dev/friel/collab-stack`
- source tip: `599ed9dc05eafd116192184bd54a2a55a2c49366`
- original base: `c1d18ceb6f22ae3acd67bbd6badad0f475b31dfc`

docs/config.md

@@ -66,6 +66,34 @@ Codex can run a notification hook when the agent finishes a turn. See the config
 
 When Codex knows which client started the turn, the legacy notify JSON payload also includes a top-level `client` field. The TUI reports `codex-tui`, and the app server reports the `clientInfo.name` value from `initialize`.
 
+## Agent Inbox Delivery
+
+By default, inbound messages from other agents are delivered to non-subagent threads as normal
+user input. If you want those handoffs to appear as explicit non-user transcript activity, you can
+opt into a synthetic function-call/function-call-output envelope:
+
+```toml
+[features]
+agent_function_call_inbox = true
+```
+
+When enabled, Codex injects inbound agent messages into non-subagent threads as an `agent_inbox`
+function-call/function-call-output pair. This is primarily a model-behavior workaround for cases
+where you want a subagent handoff to start a valid turn while still being clearly marked as
+non-user activity in the transcript.
+
+Messages sent to subagents continue to arrive as normal user input.
+
+## Watchdog Interval
+
+Watchdog agents use the top-level `watchdog_interval_s` setting to decide how long the owner thread
+must be idle before a check-in helper is spawned when the `agent_watchdog` feature is enabled.
+Owner-side typing/input activity resets the idle timer.
+
+```toml
+watchdog_interval_s = 10
+```
+
 ## JSON Schema
 
 The generated JSON Schema for `config.toml` lives at `codex-rs/core/config.schema.json`.
@@ -118,4 +146,20 @@ developer message Codex inserts when realtime becomes active. It only affects
 the realtime start message in prompt history and does not change websocket
 backend prompt settings or the realtime end/inactive message.
 
+## Custom model aliases
+
+You can add aliases to the model picker via `custom_models` in `~/.codex/config.toml`.
+Each entry maps a user-facing alias to a provider-facing model slug and can override context settings:
+
+```toml
+[[custom_models]]
+name = "gpt-5.4 1m"
+model = "gpt-5.4"
+model_context_window = 1000000
+model_auto_compact_token_limit = 900000
+```
+
+When selected, Codex sends `model = "gpt-5.4"` to the backend while using your
+alias-specific context overrides for that session.
+
 Ctrl+C/Ctrl+D quitting uses a ~1 second double-press hint (`ctrl + c again to quit`).