Cron/hook thinking-effort: downgrade only under OAuth proxy

config.example.yaml

@@ -11,25 +11,6 @@ timezone: America/New_York
 #   aws_region: us-east-1      # AWS region for Bedrock
 #   # aws_profile: ""          # Optional: AWS SSO profile name
 
-# Local proxy (CLIProxyAPI) — optional. Routes Anthropic API calls through
-# Claude Code OAuth, and is the Anthropic↔OpenAI translation layer that local
-# Ollama models are reached through. Required for the ollama block below.
-# proxy:
-#   enabled: false
-#   port: 8317
-#   host: 127.0.0.1
-
-# Local Ollama — expose locally-installed Ollama models as selectable chat
-# models in the web composer's model picker. Ollama speaks an OpenAI-compatible
-# API, so this requires proxy.enabled: true (the proxy translates the
-# Anthropic requests the SDK emits). Models are auto-discovered at runtime
-# from the running Ollama server (GET /api/tags) — whatever you've pulled
-# locally shows up automatically, no need to list models here.
-# ollama:
-#   enabled: false
-#   host: 127.0.0.1
-#   port: 11434
-
 # Agent
 agent:
   model: claude-opus-4-8         # Primary model for conversations
@@ -39,7 +20,13 @@ agent:
   title_model: claude-haiku-4-5-20251001  # Session title generation
   max_turns: 50                  # Max agentic turns per request
   max_concurrent: 4              # Max concurrent agent sessions
-  background_agent_permissions: true  # Background sub-agents (Agent run_in_background) get the same tool permissions as foreground; false denies their Write/Edit/Bash
+  thinking: max                  # Thinking budget for the main model
+  effort: max                    # Reasoning effort for the main model
+  # Note: cron/hook sessions run on cron_model (Sonnet). Under Claude OAuth
+  # (subscription) Sonnet rejects thinking/effort="max" with
+  #   level "max" not supported, valid levels: low, medium, high
+  # Nerve detects OAuth via `proxy.enabled` and automatically caps
+  # cron/hook sessions at "high" in that case. API users keep "max".
   # First-prompt rewrite — the web UI can refine the opening message of a
   # new chat with a fast model, preview it, and send only after approval.
   prompt_rewrite:

docs/config.md

@@ -40,6 +40,8 @@ from any working directory:
 | `agent.cron_model` | string | `claude-sonnet-4-6` | Model for cron jobs (cheaper) |
 | `agent.max_turns` | int | `50` | Max agentic turns per request |
 | `agent.max_concurrent` | int | `4` | Max concurrent agent sessions |
+| `agent.thinking` | string | `max` | Thinking budget for the main model: `max` / `high` / `medium` / `low` / `disabled` / `adaptive` / explicit token count. Automatically capped at `high` for cron and hook sessions when `proxy.enabled` is true (Claude OAuth subscription rejects `max` on non-flagship models like Sonnet). |
+| `agent.effort` | string | `max` | Reasoning effort for the main model: `max` / `high` / `medium` / `low`. Same OAuth+cron cap as `thinking`. |
 | `agent.prompt_rewrite.enabled` | bool | `true` | Offer the first-prompt rewrite feature in the web UI (per-user toggle lives in the composer) |
 | `agent.prompt_rewrite.model` | string | `""` | Model for prompt rewriting (empty = `agent.model`, the chat model) |
 | `agent.prompt_rewrite.max_tokens` | int | `1024` | Max tokens for the rewritten prompt |