From 363ab378afa0235afd0d5768f334add14d60f8cb Mon Sep 17 00:00:00 2001
From: Jeroen Schweitzer <jpmschweitzer@gmail.com>
Date: Thu, 18 Dec 2025 20:58:39 +0100
Subject: [PATCH] feat: optimize Housekeeper for Mistral-Nemo tool calling
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Rewrite system prompt with negative constraints and step-by-step process
- Set temperature to 0.1 for deterministic tool calling
- Sort room groups to top of device list (address positional bias)
- Add [ROOM GROUP] marker in list_devices output
- Update tool docstrings with explicit entity_id= parameter examples
- Add optimization findings doc (experiment log: 0% → 100% success)
- Add test script for room group detection regression testing

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---
 CHANGELOG.md                              |  12 ++
 docs/housekeeper-optimization-findings.md | 246 ++++++++++++++++++++++
 scripts/test_housekeeper.sh               | 141 +++++++++++++
 src/agents/housekeeper/agent.py           | 126 +++++------
 src/agents/housekeeper/tools.py           |  49 +++--
 5 files changed, 488 insertions(+), 86 deletions(-)
 create mode 100644 docs/housekeeper-optimization-findings.md
 create mode 100755 scripts/test_housekeeper.sh

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 33862ce..f5fcb15 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Changed
+
+- **Housekeeper prompt optimization** - Rewrote system prompt for Mistral-Nemo function calling with negative constraints, step-by-step process, and explicit entity ID format guidance
+- **Housekeeper temperature setting** - Set temperature to 0.1 for deterministic tool calling behavior
+- **Device list room group priority** - Room groups now appear first in `list_devices` output with `[ROOM GROUP]` marker to address positional bias
+- **Tool docstring improvements** - Updated turn_on/turn_off/toggle with explicit `entity_id=` parameter examples
+
+### Added
+
+- **Housekeeper optimization findings** - Added `docs/housekeeper-optimization-findings.md` documenting the experiment journey from 0% to 100% success rate
+- **Housekeeper test script** - Added `scripts/test_housekeeper.sh` for room group detection regression testing
+
 ## [1.8.6] - 2025-12-17
 
 ### Fixed
diff --git a/docs/housekeeper-optimization-findings.md b/docs/housekeeper-optimization-findings.md
new file mode 100644
index 0000000..286862b
--- /dev/null
+++ b/docs/housekeeper-optimization-findings.md
@@ -0,0 +1,246 @@
+# Housekeeper Agent Optimization Findings
+
+## Background
+
+Research with Gemini identified key issues with mistral-nemo and tool calling:
+- "Pre-computation Hallucination" - model answers before using tools
+- High default temperature (0.7-0.8) causes wandering
+- Model is "chatty and confident" - needs explicit constraints
+
+## Key Recommendations from Gemini Research
+
+1. **Temperature 0.0** for tool-calling agents (deterministic, follows schema)
+2. **Chain of Thought (CoT)** - force step-by-step reasoning
+3. **Negative constraints** - tell model what NOT to do (Nemo responds better)
+4. **Explicit tool descriptions** - verbose docstrings with "never estimate yourself"
+5. **"Strictly tool-based assistant"** pattern - NO internal knowledge claim
+
+---
+
+## Experiment Log
+
+### Baseline (v1.8.6)
+- **Date**: 2025-12-17
+- **Configuration**: Default temperature, improved prompt requiring list_devices first
+- **Results**:
+  - Called list_devices first ✓
+  - Still hallucinated `light.study_desk` despite seeing list with only `light.study` and `light.study_main`
+  - Partial success: turned off `light.study_main`, failed on hallucinated entity
+- **Success rate**: ~50% (1 of 2 study lights controlled correctly)
+
+---
+
+### Experiment 1: Temperature 0.0
+- **Date**: 2025-12-18
+- **Change**: Set `model_settings=ModelSettings(temperature=0.0)` for Housekeeper
+- **Hypothesis**: Deterministic output will force model to use exact entity IDs from tool results
+- **Results**:
+
+  **Study lights test:**
+  - Called `list_devices()` first ✓ (but no domain filter)
+  - Used wrong parameter `device_id` instead of `entity_id` (recovered after validation error)
+  - Only identified `light.studeerlamp` as "study" related (Dutch name)
+  - **Missed `light.study` and `light.study_main`** - didn't match English "study"
+  - Turned off 1 wrong light, missed 2 actual study lights
+
+  **Kitchen lights test:**
+  - Called `list_devices()` first ✓ (no domain filter)
+  - Saw full device list including `light.kitchen`
+  - Used wrong parameter `device_id` instead of `entity_id` (recovered after validation)
+  - After correction, dropped domain prefix: used `kitchen` instead of `light.kitchen`
+  - 404 error - device not found
+
+- **Success rate**: 0% (no target lights successfully controlled)
+- **Observations**:
+  - Temperature 0.0 alone is insufficient
+  - Model consistently confuses `device_id` vs `entity_id` parameter name
+  - After validation error correction, model truncates entity_id (drops domain prefix)
+  - Semantic matching of room names to devices is weak
+  - Model doesn't understand entity_id format: `domain.name`
+
+---
+
+### Experiment 2: Negative Constraints + CoT
+- **Date**: 2025-12-18
+- **Change**: Complete prompt rewrite with:
+  - "You have NO Internal Knowledge" - negative framing
+  - Explicit entity_id format with WRONG/RIGHT examples
+  - Step-by-step process (ALWAYS FOLLOW)
+  - Explicit parameter names section
+  - "What NOT To Do" negative constraints
+- **Hypothesis**: Negative constraints work better with Mistral-Nemo
+- **Results**:
+
+  **Study lights test:**
+  - Called `list_devices(domain="light")` ✓ with domain filter (improvement!)
+  - Still used `device_id` first, recovered to `entity_id` after validation error
+  - After recovery, used correct full format: `light.studeerlamp`
+  - **Still only matched `studeerlamp` not `light.study` or `light.study_main`**
+
+  **Kitchen lights test:**
+  - Called `list_devices(domain="light")` ✓
+  - Called `turn_off(entity_id="light.kitchen")` ✓ correct format!
+  - All 4 kitchen lights turned off (light.kitchen is a group)
+  - **100% success for kitchen!**
+
+- **Success rate**:
+  - Study: 0% (wrong semantic match)
+  - Kitchen: 100% (4/4 lights off)
+  - Combined: ~50% (1 of 2 tests successful)
+- **Observations**:
+  - Domain filter now consistently used ✓
+  - Entity_id format correct after recovery ✓
+  - Semantic matching still fails for "study" → prefers Dutch "studeerlamp" over English "study"
+  - Parameter name confusion persists (`device_id` vs `entity_id`)
+  - Simple room names (kitchen) work; mixed language fails (study/studeerlamp)
+
+---
+
+### Experiment 3: Temperature 0.1 + Explicit Tool Docstrings
+- **Date**: 2025-12-18
+- **Change**:
+  - Temperature 0.1
+  - Updated turn_on/turn_off docstrings with explicit `entity_id=` in examples
+- **Results**:
+  - Still uses `device_id` first, recovers to `entity_id` after validation
+  - Still picks wrong entity (studeerlamp over study)
+- **Success rate**: 0%
+
+---
+
+### Experiment 4: Room Group Priority (with explicit examples)
+- **Date**: 2025-12-18
+- **Change**: Updated prompt with:
+  - Explicit instruction: "Look for EXACT match `light.<room_name>` first!"
+  - Concrete examples: "For 'study lights' → look for `light.study`"
+  - Working example showing `turn_off(entity_id="light.study")`
+- **Hypothesis**: Explicit examples will guide model to use room groups
+- **Results**:
+
+  **Test 1 & 2 (consecutive):**
+  - Called `list_devices(domain="light")` ✓
+  - Device list clearly shows `light.study` at the bottom
+  - First call: `turn_off({"devices":["studeerlamp"]})` - wrong param AND wrong device
+  - After validation error: `turn_off(entity_id="light.studeerlamp")` - correct param, still wrong device
+  - **Completely ignored `light.study` despite prompt explicitly saying to use it**
+
+- **Success rate**: 0% (wrong device controlled)
+- **Observations**:
+  - Model ignores explicit step-by-step instructions in favor of substring matching
+  - Dutch "studeerlamp" contains "studer" which the model prefers over exact "study" match
+  - Even when prompt has a literal example `turn_off(entity_id="light.study")`, model uses `light.studeerlamp`
+  - Positional bias possible - `light.study` appears at end of 21-item list
+  - **Fundamental limitation**: Mistral-Nemo cannot follow explicit matching rules
+
+---
+
+### Experiment 5: Room Groups First (Tool Output Ordering)
+- **Date**: 2025-12-18
+- **Change**: Modified `list_devices` to sort room groups to top of list using HA attributes (`is_hue_group`, `hue_type="room"`)
+- **Hypothesis**: Positional bias - model focuses on items earlier in list
+- **Results**:
+  - Room groups (`light.study`, `light.kitchen`, etc.) now appear first in device list
+  - Combined with improved prompt, model now consistently uses room groups
+  - **70% success rate** (7/10 tests) with default q4 quantization
+
+---
+
+### Experiment 6: Model Quantization (q5_1)
+- **Date**: 2025-12-18
+- **Change**: Upgraded from default Mistral-Nemo quantization (q4) to `mistral-nemo:12b-instruct-2407-q5_1`
+- **Hypothesis**: Higher precision weights improve tool calling accuracy
+- **Results**:
+
+  | Test | Action | Result |
+  |------|--------|--------|
+  | 1 | Turn off study | PASS |
+  | 2 | Turn on study | PASS |
+  | 3 | Toggle study | PASS |
+  | 4 | Turn off kitchen | PASS |
+  | 5 | Turn on kitchen | PASS |
+  | 6 | Toggle kitchen | PASS |
+  | 7 | Turn off bedroom | PASS |
+  | 8 | Turn on bedroom | PASS |
+  | 9 | Turn off living room | PASS |
+  | 10 | Turn on living room | PASS |
+
+- **Success rate**: **100%** (10/10 tests)
+- **Observations**:
+  - q5_1 quantization dramatically improves tool calling accuracy
+  - All room groups correctly identified and used
+  - No parameter confusion (`entity_id` used correctly)
+  - No entity_id truncation issues
+  - Toggle operations now work reliably
+  - Model fits within 10GB VRAM (q6 did not)
+
+---
+
+### Experiment 7: Device List in System Prompt (Context Injection)
+- **Date**: [PENDING]
+- **Change**: Store device list in database (per user/household) and inject into system prompt
+- **Approach**:
+  1. Periodically sync device list from Home Assistant to PostgreSQL
+  2. On each Housekeeper invocation, fetch device list and include in prompt
+  3. Remove need for model to call list_devices() - just match from context
+- **Hypothesis**:
+  - Eliminates tool call step where errors occur
+  - Reduces context size by not returning full device list as tool output
+  - Makes entity matching a language task (in prompt) rather than tool result parsing
+- **Trade-offs**:
+  - Stale data if sync is infrequent
+  - Prompt size increase (but less than tool call response)
+  - Need sync mechanism and storage
+- **Results**: [TO BE RECORDED]
+- **Success rate**: [TO BE RECORDED]
+
+---
+
+## Key Problem Identified (Solved)
+
+The model struggled with:
+1. **Parameter schema adherence** - uses `device_id` when schema requires `entity_id`
+2. **Value preservation** - truncates values after validation errors (drops `light.` prefix)
+3. **Semantic matching** - prefers substring matches ("studeerlamp" contains "studer") over exact matches (`light.study`)
+4. **Following explicit instructions** - ignores step-by-step processes even when examples are provided
+5. **Positional bias** - may not "see" items at the end of long lists
+
+**Solution**: These issues were resolved by:
+1. Using q5_1 quantization instead of default q4 (higher precision weights)
+2. Sorting room groups to top of device list (address positional bias)
+3. Explicit prompt guidance with negative constraints and examples
+
+---
+
+## Potential Next Experiments
+
+### Experiment 5: Room Groups First (List Ordering)
+- **Hypothesis**: Positional bias - model focuses on items earlier in list
+- **Change**: Sort device list to put room groups (entities matching `light.<single_word>`) at the TOP
+- **Effort**: Low - modify list_devices output formatting
+- **Risk**: May affect other use cases where individual devices are needed
+
+### Experiment 6: Simplified Device List Format
+- **Hypothesis**: Markdown formatting adds noise that confuses the model
+- **Change**: Return simple list: `light.study (Study - GROUP), light.study_main (Ceiling light), ...`
+- **Effort**: Low - modify list_devices output
+- **Risk**: Less human-readable responses
+
+---
+
+## Learnings to Apply Elsewhere
+
+1. **Quantization matters** - q5_1 dramatically outperforms q4 for tool calling (100% vs 70%)
+2. **Positional bias is real** - sort important items to top of lists
+3. **Smaller models need simpler workflows** - fewer tool calls, more context injection
+4. **Validation errors don't teach** - model often makes worse mistakes on retry
+5. **Entity IDs are hard** - domain.name format confuses the model
+6. **Consider pre-computation** - move matching logic to code, not LLM
+7. **Use explicit negative constraints** - "NEVER do X" works better than "always do Y"
+
+---
+
+## Notes
+
+- Librarian may need higher temperature for creative synthesis
+- All "action" agents (Housekeeper, future agents) should use low temperature
+- Consider testing with Gemma 2 9B for better function calling (Google, open weights)
diff --git a/scripts/test_housekeeper.sh b/scripts/test_housekeeper.sh
new file mode 100755
index 0000000..1efcb4c
--- /dev/null
+++ b/scripts/test_housekeeper.sh
@@ -0,0 +1,141 @@
+#!/bin/bash
+# Housekeeper Room Group Detection Test Suite
+# Verifies room groups are controlled by checking actual state changes
+
+API_URL="http://localhost:8777/v1/chat/completions"
+CORE_API="http://192.168.86.149:8083"
+RESULTS_FILE="/tmp/housekeeper_test_results.txt"
+
+GREEN='\033[0;32m'
+RED='\033[0;31m'
+YELLOW='\033[1;33m'
+NC='\033[0m'
+
+get_state() {
+    curl -s "$CORE_API/housekeeping/devices/$1" 2>/dev/null | jq -r '.state' 2>/dev/null
+}
+
+echo "=========================================="
+echo "Housekeeper Room Group Test Suite"
+echo "=========================================="
+echo ""
+
+> "$RESULTS_FILE"
+
+run_toggle_test() {
+    local test_num=$1
+    local room=$2
+    local entity="light.$room"
+    local prompt_room="${room//_/ }"
+
+    printf "Test %2d: Toggle %-12s lights ... " "$test_num" "$prompt_room"
+
+    local before=$(get_state "$entity")
+    if [ -z "$before" ] || [ "$before" = "null" ]; then
+        echo -e "${YELLOW}SKIP${NC} (cannot get state)"
+        echo "SKIP|$test_num|Toggle $room|error" >> "$RESULTS_FILE"
+        return
+    fi
+
+    curl -s -X POST "$API_URL" \
+        -H "Content-Type: application/json" \
+        -d "{\"model\": \"tatlock\", \"messages\": [{\"role\": \"user\", \"content\": \"Toggle the $prompt_room lights\"}]}" > /dev/null
+
+    sleep 4
+
+    local after=$(get_state "$entity")
+
+    if [ "$before" != "$after" ]; then
+        echo -e "${GREEN}PASS${NC} ($before -> $after)"
+        echo "PASS|$test_num|Toggle $room|$before->$after" >> "$RESULTS_FILE"
+    else
+        echo -e "${RED}FAIL${NC} (state unchanged: $before)"
+        echo "FAIL|$test_num|Toggle $room|unchanged:$before" >> "$RESULTS_FILE"
+    fi
+}
+
+run_onoff_test() {
+    local test_num=$1
+    local room=$2
+    local action=$3
+    local expected_state=$4
+    # Entity uses underscore, prompt uses space
+    local entity="light.${room//_/ }"
+    entity="light.$room"
+    local prompt_room="${room//_/ }"
+
+    printf "Test %2d: %-8s %-12s lights ... " "$test_num" "$action" "$prompt_room"
+
+    curl -s -X POST "$API_URL" \
+        -H "Content-Type: application/json" \
+        -d "{\"model\": \"tatlock\", \"messages\": [{\"role\": \"user\", \"content\": \"$action the $prompt_room lights\"}]}" > /dev/null
+
+    sleep 4
+
+    local after=$(get_state "$entity")
+
+    if [ "$after" = "$expected_state" ]; then
+        echo -e "${GREEN}PASS${NC} ($after)"
+        echo "PASS|$test_num|$action $room|$after" >> "$RESULTS_FILE"
+    else
+        echo -e "${RED}FAIL${NC} (got $after, expected $expected_state)"
+        echo "FAIL|$test_num|$action $room|got:$after,expected:$expected_state" >> "$RESULTS_FILE"
+    fi
+}
+
+echo "Running tests (~4s each)..."
+echo ""
+
+# Study tests
+run_onoff_test 1 "study" "Turn off" "off"
+run_onoff_test 2 "study" "Turn on" "on"
+run_toggle_test 3 "study"
+
+# Kitchen tests
+run_onoff_test 4 "kitchen" "Turn off" "off"
+run_onoff_test 5 "kitchen" "Turn on" "on"
+run_toggle_test 6 "kitchen"
+
+# Bedroom tests
+run_onoff_test 7 "bedroom" "Turn off" "off"
+run_onoff_test 8 "bedroom" "Turn on" "on"
+
+# Living room tests (entity is light.living_room)
+run_onoff_test 9 "living_room" "Turn off" "off"
+run_onoff_test 10 "living_room" "Turn on" "on"
+
+# Ensure all lights end up ON
+echo ""
+echo "Restoring all lights to ON..."
+for room in "study" "kitchen" "bedroom" "living room"; do
+    curl -s -X POST "$API_URL" \
+        -H "Content-Type: application/json" \
+        -d "{\"model\": \"tatlock\", \"messages\": [{\"role\": \"user\", \"content\": \"Turn on the $room lights\"}]}" > /dev/null
+    sleep 3
+done
+echo "Done."
+
+echo ""
+echo "=========================================="
+echo "Results"
+echo "=========================================="
+
+PASS=$(grep -c "^PASS" "$RESULTS_FILE" 2>/dev/null || echo 0)
+FAIL=$(grep -c "^FAIL" "$RESULTS_FILE" 2>/dev/null || echo 0)
+SKIP=$(grep -c "^SKIP" "$RESULTS_FILE" 2>/dev/null || echo 0)
+TOTAL=$((PASS + FAIL))
+
+echo "Passed:  $PASS"
+echo "Failed:  $FAIL"
+echo "Skipped: $SKIP"
+
+if [ "$TOTAL" -gt 0 ]; then
+    echo ""
+    echo "Success Rate: $((PASS * 100 / TOTAL))% ($PASS/$TOTAL)"
+fi
+
+if [ "$FAIL" -gt 0 ]; then
+    echo ""
+    echo "Failures:"
+    grep "^FAIL" "$RESULTS_FILE"
+fi
diff --git a/src/agents/housekeeper/agent.py b/src/agents/housekeeper/agent.py
index cef931a..4870a64 100644
--- a/src/agents/housekeeper/agent.py
+++ b/src/agents/housekeeper/agent.py
@@ -32,93 +32,69 @@ from src.core.logging_config import get_logger
 
 logger = get_logger(__name__)
 
-# Housekeeper system prompt
-HOUSEKEEPER_SYSTEM_PROMPT = """You are The Housekeeper, an expert home automation assistant in the Tatlock household.
+# Housekeeper system prompt - Optimized for Mistral-Nemo function calling
+HOUSEKEEPER_SYSTEM_PROMPT = """You are a strictly tool-based home automation assistant.
 
-Your role is to help users control and monitor their smart home through Home Assistant:
-- Lights, switches, and other devices
-- Scenes (pre-configured device states)
-- Scripts (automation sequences)
-- Automations (event-triggered rules)
+## CRITICAL: You Have NO Internal Knowledge
 
-## Your Personality
-- Efficient and practical
-- Safety-conscious (confirm destructive actions)
-- Proactive in suggesting optimizations
-- Clear about what actions you're taking
+You do NOT know what devices exist. You do NOT know any entity IDs.
+Entity IDs are different in every installation. You MUST discover them using tools.
 
-## Your Tools
+## Entity ID Format
 
-### Discovery Tools
-- **list_areas**: See all rooms/areas configured in Home Assistant
-- **list_devices**: Find devices by type (domain) or location (area)
-- **get_device_state**: Check a device's current state and attributes
+Entity IDs follow the format: `domain.name`
+Examples: `light.kitchen`, `light.study_main`, `switch.coffee_maker`
 
-### Control Tools
-- **turn_on**: Turn on lights, switches, etc. (supports brightness/color for lights)
-- **turn_off**: Turn off devices
-- **toggle**: Flip a device's state
+The `entity_id` parameter MUST be the COMPLETE value including the domain prefix.
+WRONG: `entity_id="kitchen"`
+RIGHT: `entity_id="light.kitchen"`
 
-### Scene Tools
-- **list_scenes**: See available scene presets
-- **activate_scene**: Activate a scene (e.g., "movie night", "good morning")
+## Step-by-Step Process (ALWAYS FOLLOW)
 
-### Script Tools
-- **list_scripts**: See available automation scripts
-- **run_script**: Execute a script
+When asked to control devices in a room:
 
-### Automation Tools
-- **list_automations**: See all automations and their status
-- **toggle_automation**: Enable or disable an automation
+1. THINK: What domain? (light, switch, climate, etc.)
+2. CALL: list_devices(domain="light") to discover available devices
+3. CHECK: Look for EXACT match `light.<room_name>` first!
+   - For "study lights" → look for `light.study` (not light.study_main, not light.studeerlamp)
+   - For "kitchen lights" → look for `light.kitchen` (not light.kitchen_spot_1)
+   - These room groups control ALL lights in that room at once
+   - If found, use ONLY the group (stop looking for individual lights)
+4. FALLBACK: Only if no exact room group exists, find entity_ids containing the room name
+5. CALL: turn_on/turn_off using the EXACT entity_id from step 3 or 4
 
-### History Tools
-- **get_history**: Check a device's state history
+Example for "Turn off study lights":
+1. Domain is "light"
+2. Call list_devices(domain="light")
+3. Look for room group: `light.study` - FOUND!
+4. Call turn_off(entity_id="light.study")  # This controls all study lights
 
-## Critical Rules
+Example for "Turn off hallway lights" (no room group):
+1. Domain is "light"
+2. Call list_devices(domain="light")
+3. Look for room group: `light.hallway` - NOT FOUND
+4. Find all with "hallway": light.hallway_spot_1, light.hallway_spot_2
+5. Call turn_off for each
 
-**NEVER GUESS ENTITY IDs.** You do not know what devices exist. Entity IDs vary between
-installations. You MUST call list_devices() FIRST to discover actual entity_ids before
-ANY control action (turn_on, turn_off, toggle).
+## Tool Parameter Names
 
-Wrong approach:
-  User: "Turn off the study lights"
-  You: turn_off("light.study")  ← WRONG! You guessed the entity_id
+- turn_on, turn_off, toggle: Use `entity_id` (NOT device_id, NOT id)
+- activate_scene: Use `scene_id`
+- run_script: Use `script_id`
 
-Correct approach:
-  User: "Turn off the study lights"
-  You: list_devices(domain="light")  ← First discover what exists
-  You: [See results like light.study_main, light.study]
-  You: turn_off("light.study_main"), turn_off("light.study")  ← Use actual IDs
+## What NOT To Do
 
-## Best Practices
-
-1. **ALWAYS list_devices first** before any control action. No exceptions.
-   Filter by domain and/or area to narrow results.
-
-2. **Use exact entity_ids** from list_devices results. Never construct or guess them.
-
-3. **Area-aware filtering**: Use area parameter when users mention a room.
-   Note: Some devices may have area=None but contain the room name in entity_id.
-
-4. **Verify after actions**: Use get_device_state to confirm state changes if needed.
-
-5. **Safety for bulk actions**: When affecting multiple devices, summarize first.
-
-## Common Patterns
-
-- "Turn on the lights" → list_devices(domain="light"), then turn_on each
-- "What's on?" → list_devices() and filter for state="on"
-- "Movie time" → Either activate_scene("scene.movie_night") or run_script if available
-- "Dim the bedroom" → turn_on("light.bedroom", brightness=64)
+- NEVER guess an entity_id
+- NEVER construct an entity_id from the room name
+- NEVER drop the domain prefix (light., switch., etc.)
+- NEVER use "device_id" - the parameter is called "entity_id"
+- NEVER provide an answer without calling list_devices first
 
 ## Response Format
-Your responses are returned to Tatlock (the butler) who will synthesize them into
-a final answer for the user. Keep this in mind:
-- Lead with confirmation of what you did or found
-- Be specific about which devices were affected
-- Include relevant state information
-- Note any issues or failures
-- Be concise - Tatlock will format the final response
+
+After completing actions, briefly confirm:
+- Which devices were affected (list the entity_ids)
+- Whether each action succeeded or failed
 """
 
 # Lazy initialization to avoid connection issues during imports
@@ -231,9 +207,13 @@ async def run_housekeeper(
     )
 
     try:
+        # Use temperature 0.1 for slight exploration
+        from pydantic_ai.settings import ModelSettings
+
         result = await agent.run(
             prompt,
             message_history=message_history,
+            model_settings=ModelSettings(temperature=0.1),
         )
 
         logger.info(
@@ -289,9 +269,13 @@ async def run_housekeeper_stream(
     )
 
     try:
+        # Use temperature 0.1 for slight exploration
+        from pydantic_ai.settings import ModelSettings
+
         async with agent.run_stream(
             prompt,
             message_history=message_history,
+            model_settings=ModelSettings(temperature=0.1),
         ) as response:
             async for delta in response.stream_text(delta=True):
                 yield delta
diff --git a/src/agents/housekeeper/tools.py b/src/agents/housekeeper/tools.py
index cf8d857..e95946d 100644
--- a/src/agents/housekeeper/tools.py
+++ b/src/agents/housekeeper/tools.py
@@ -59,10 +59,27 @@ async def list_devices(
 
             for dom, dom_devices in sorted(by_domain.items()):
                 output_parts.append(f"### {dom.title()}s")
-                for device in dom_devices:
+
+                # Sort devices: room groups first (using Home Assistant's is_hue_group attribute)
+                def is_room_group(d: object) -> bool:
+                    """Check if device is a room group based on HA attributes."""
+                    attrs = getattr(d, "attributes", {})
+                    # Check for Hue room groups
+                    if attrs.get("is_hue_group") and attrs.get("hue_type") == "room":
+                        return True
+                    # Check for other group indicators (icon or entity_id list)
+                    if "entity_id" in attrs and isinstance(attrs["entity_id"], list):
+                        return True
+                    return False
+
+                sorted_devices = sorted(dom_devices, key=lambda d: (not is_room_group(d), d.entity_id))
+
+                for device in sorted_devices:
                     state_icon = "on" if device.state == "on" else "off" if device.state == "off" else device.state
                     area_str = f" ({device.area})" if device.area else ""
-                    output_parts.append(f"- **{device.name}**{area_str}: {state_icon}")
+                    # Mark room groups clearly using actual HA data
+                    group_marker = " [ROOM GROUP]" if is_room_group(device) else ""
+                    output_parts.append(f"- **{device.name}**{area_str}{group_marker}: {state_icon}")
                     output_parts.append(f"  ID: `{device.entity_id}`")
                 output_parts.append("")
 
@@ -164,12 +181,12 @@ async def turn_on(
     color_temp: int | None = None,
 ) -> str:
     """
-    Turn on a device.
+    Turn on a device. Use the entity_id parameter with the EXACT value from list_devices.
 
     For lights, can optionally set brightness and color temperature.
 
     Args:
-        entity_id: Device to turn on (e.g., light.living_room, switch.coffee_maker)
+        entity_id: The EXACT entity ID from list_devices including domain prefix.
         brightness: Optional brightness for lights (0-255, where 255 is full brightness)
         color_temp: Optional color temperature in Kelvin (2700=warm, 6500=cool)
 
@@ -177,10 +194,9 @@ async def turn_on(
         Confirmation of the action
 
     Examples:
-        turn_on("light.living_room")  # Turn on at current brightness
-        turn_on("light.bedroom", brightness=128)  # Turn on at 50% brightness
-        turn_on("light.office", brightness=255, color_temp=4000)  # Full, neutral white
-        turn_on("switch.coffee_maker")  # Turn on a switch
+        turn_on(entity_id="light.living_room")
+        turn_on(entity_id="light.bedroom", brightness=128)
+        turn_on(entity_id="switch.coffee_maker")
     """
     try:
         async with CoreAPIClient() as client:
@@ -209,17 +225,18 @@ async def turn_on(
 
 async def turn_off(entity_id: str) -> str:
     """
-    Turn off a device.
+    Turn off a device. Use the entity_id parameter with the EXACT value from list_devices.
 
     Args:
-        entity_id: Device to turn off (e.g., light.living_room, switch.coffee_maker)
+        entity_id: The EXACT entity ID from list_devices including domain prefix.
 
     Returns:
         Confirmation of the action
 
     Examples:
-        turn_off("light.living_room")
-        turn_off("switch.coffee_maker")
+        turn_off(entity_id="light.living_room")
+        turn_off(entity_id="switch.coffee_maker")
+        turn_off(entity_id="light.kitchen")
     """
     try:
         async with CoreAPIClient() as client:
@@ -239,15 +256,17 @@ async def toggle(entity_id: str) -> str:
     """
     Toggle a device's state (on becomes off, off becomes on).
 
+    Use the entity_id parameter with the EXACT value from list_devices.
+
     Args:
-        entity_id: Device to toggle
+        entity_id: The EXACT entity ID from list_devices including domain prefix.
 
     Returns:
         Confirmation with the new state
 
     Examples:
-        toggle("light.living_room")
-        toggle("switch.fan")
+        toggle(entity_id="light.living_room")
+        toggle(entity_id="switch.fan")
     """
     try:
         async with CoreAPIClient() as client: