fix(paseo-agent): retain prompts and required tools

Emit submitted prompts into the daemon timeline so refresh and reload can replay the full Paseo Agent conversation. Keep nonpersistent providers from getting synthetic resume handles, and gate Paseo tool injection through provider capabilities.

docs/paseo-agent.md

@@ -19,6 +19,15 @@ instance and the `openai/gpt-4o-mini` model exposed by that instance.
 > `--host <addr>` to CLI smoke commands so they hit your isolated daemon, not the real
 > daemon on `:6767`.
 
+## Timeline and persistence
+
+Paseo Agent does not write or resume Pi CLI session files such as
+`~/.pi/agent/sessions/*`; it runs Pi's harness in process and currently reports no
+provider-owned session persistence. App refreshes and daemon-side agent reloads therefore
+depend on Paseo's own timeline rows, not on a Pi history replay. Submitted prompts must be
+emitted as `user_message` timeline rows when the turn starts; assistant text, reasoning,
+and tool calls arrive later from Pi events and are mapped into the same Paseo timeline.
+
 ## Provider catalog
 
 Paseo Agent model providers are catalog-driven, but Paseo does not copy Pi's provider
@@ -255,6 +264,9 @@ Transports: HTTP (streamable) is the primary path (the injected `paseo` server i
 SSE and stdio transports are also wired via the MCP SDK. No extra config is needed: MCP
 servers come from Paseo's normal injection/config, not from `agents.paseo`.
 
+The internal `paseo` MCP server is required for Paseo Agent sessions and is injected even
+when the global `mcp.injectIntoAgents` setting is disabled for other providers.
+
 ## Agent definitions
 
 Paseo Agent can load a Paseo-owned agent definition from `$PASEO_HOME/agents/*.md`.