Self-Improving Domotics Skill

Never overwrite existing files. The commands are no-op when already initialised.

Do not log secrets, access tokens, lock PINs, alarm codes, network credentials, or personal household schedules in plain text.

If automatic reminders are needed, enable hook integration (opt-in) in Hook Integration.

Quick Reference

OpenClaw Setup (Recommended)

OpenClaw is the primary platform for this skill. It supports workspace injection and optional hooks for proactive reminders.

Installation

Via ClawdHub (recommended):

clawdhub install self-improving-domotics

Manual clone:

git clone https://github.com/jose-compu/self-improving-domotics.git ~/.openclaw/skills/self-improving-domotics

Workspace Structure

OpenClaw injects workspace files at session start:

~/.openclaw/workspace/
├── AGENTS.md / SOUL.md / TOOLS.md / MEMORY.md
├── memory/YYYY-MM-DD.md
└── .learnings/{LEARNINGS.md, DOMOTICS_ISSUES.md, FEATURE_REQUESTS.md}

Create Learning Files

mkdir -p ~/.openclaw/workspace/.learnings

Then create (or copy from assets/) the domotics logs:

• LEARNINGS.md - reliability and optimization learnings
• DOMOTICS_ISSUES.md - concrete failures and incident-style timelines
• FEATURE_REQUESTS.md - automation and observability capability requests

Promotion Targets

Promote proven learnings to durable assets:

Optional: Enable Hook

For reminder injection during session bootstrap:

cp -r hooks/openclaw ~/.openclaw/hooks/self-improving-domotics
openclaw hooks enable self-improving-domotics

See references/openclaw-integration.md for details.

Generic Setup (Other Agents)

For Claude Code, Codex, Copilot, or other agents:

mkdir -p .learnings

Create the three files with the headers shown in First-Use Initialisation.

Add reference to agent instruction files

Add to AGENTS.md, CLAUDE.md, or .github/copilot-instructions.md:

Self-Improving Domotics Workflow

When smart-home issues or patterns are discovered:

1. Log to .learnings/DOMOTICS_ISSUES.md, .learnings/LEARNINGS.md, or .learnings/FEATURE_REQUESTS.md
2. Promote verified recurring patterns to:
   • Home automation playbooks
   • Device compatibility matrix
   • Automation rule library
   • Safety automations

Logging Format

Learning Entry [LRN-YYYYMMDD-XXX]

Append to .learnings/LEARNINGS.md:

## [LRN-YYYYMMDD-XXX] category

**Logged**: ISO-8601 timestamp
**Priority**: low | medium | high | critical
**Status**: pending
**Area**: lighting | climate | security | access_control | energy | sensors | routines | voice_assistants | integrations | network

### Summary
One-line domotics learning

### Details
What happened, why behavior was suboptimal, what rule/device/integration assumptions failed,
and what robust approach should be preferred.

### Safe Operating Notes
- High-impact actuator involved: yes | no
- Human confirmation required: yes | no
- Fallback on uncertainty: no action | notify-only | safe mode

### Suggested Action
Specific design, rule, configuration, or documentation change

### Metadata
- Source: automation_event | incident_review | user_report | monitoring_alert | simplify-and-harden
- Category: automation_conflict | sensor_drift | device_unreachable | integration_break | energy_optimization | safety_rule_gap | occupancy_mismatch | latency_jitter
- Related Devices: thermostat-v2, zigbee-motion-3
- Related Integrations: home-assistant | mqtt | matter | zigbee2mqtt | alexa
- Related Entries: DOM-20260413-001
- Tags: tag1, tag2
- See Also: LRN-20260410-002
- Pattern-Key: occupancy.false_positive_night | integration.schema_change (optional)
- Recurrence-Count: 1 (optional)
- First-Seen: 2026-04-13 (optional)
- Last-Seen: 2026-04-13 (optional)

---

Domotics Issue Entry [DOM-YYYYMMDD-XXX]

Append to .learnings/DOMOTICS_ISSUES.md:

## [DOM-YYYYMMDD-XXX] issue_name

**Logged**: ISO-8601 timestamp
**Priority**: medium | high | critical
**Status**: pending
**Area**: lighting | climate | security | access_control | energy | sensors | routines | voice_assistants | integrations | network

### Summary
Brief issue statement

### Incident Timeline
| Time | Event |
|------|-------|
| T+0m | Trigger observed |
| T+Xm | Detection or alert |
| T+Xm | Mitigation applied |
| T+Xm | Verified stable or still failing |

### Root Cause
Underlying reason (rule conflict, stale sensor, network partition, vendor API change, etc.)

### Correct Approach
How system design or operating procedure should handle this scenario

### Home Impact
- User Impact: none | minor | significant
- Safety Impact: none | potential | high
- Security Impact: none | potential | high
- Energy Impact: neutral | increased | reduced
- Manual Intervention: none | required

### Prevention
Guardrails, retries, validation, fallback policy, notification policy, or routine redesign

### Context
- Trigger: monitoring_alert | user_report | regression | integration_update | schedule_miss
- Topology: local_only | cloud_assisted | hybrid
- Network Layer: wifi | zigbee | zwave | thread | ethernet | mixed

### Metadata
- Related Devices: lock-front-door, motion-hallway
- Related Files: automations/night-routine.yaml
- Related Entries: LRN-20260413-001
- See Also: DOM-20260410-002

---

Feature Request Entry [FEAT-YYYYMMDD-XXX]

Append to .learnings/FEATURE_REQUESTS.md:

## [FEAT-YYYYMMDD-XXX] capability_name

**Logged**: ISO-8601 timestamp
**Priority**: low | medium | high
**Status**: pending
**Area**: lighting | climate | security | access_control | energy | sensors | routines | voice_assistants | integrations | network

### Requested Capability
What automation or reliability capability is needed

### User Context
Why this matters, what failure pattern it prevents, what experience it improves

### Complexity Estimate
simple | medium | complex

### Suggested Implementation
Specific architecture, tooling, API contracts, and rollout safety controls

### Metadata
- Frequency: first_time | recurring
- Related Features: feature_name_or_component

---

Entry Types and Categories

Entry Types

Categories

ID Generation

Format: TYPE-YYYYMMDD-XXX

• TYPE: LRN | DOM | FEAT
• YYYYMMDD: current date
• XXX: sequential or random 3-char suffix (001, A9K)

Examples:

• LRN-20260413-001
• DOM-20260413-A7B
• FEAT-20260413-002

Resolving Entries

When addressed, update:

1. **Status**: pending -> **Status**: resolved
2. Add resolution section:

### Resolution
- **Resolved**: 2026-04-13T18:20:00Z
- **Action Taken**: Added cooldown and guard condition to occupancy-lighting routine
- **Notes**: Added manual confirmation before lock and alarm state transitions at night

Other valid statuses:

• in_progress - actively investigated
• wont_fix - accepted as non-actionable (include reason)
• promoted - promoted to permanent target
• promoted_to_skill - extracted to reusable skill

Detection Triggers

Automatically log when these domotics signals appear.

Automation Conflict Triggers (-> DOM or LRN)

• Rapid repeated toggles, opposing commands, or circular rule loops
• Overlapping conditions without explicit precedence

Device Reachability Triggers (-> DOM)

• Offline/unreachable warnings, repeated timeouts, missing acknowledgments
• Zigbee/Z-Wave mesh drops or MQTT entity unavailable states

Integration Break Triggers (-> DOM or LRN)

• Webhook/API auth/schema/endpoint changes causing parser or action failures
• Voice-assistant intent mapping regressions after platform updates

Occupancy and Sensor Triggers (-> LRN)

• Occupancy confidence mismatches observed reality
• Stale/impossible sensor values beyond threshold windows

Timing and Schedule Triggers (-> DOM)

• Missed schedules, delayed execution, or jitter-induced duplicate actions
• Timezone/sunrise-sunset offset errors affecting routines

Safety Triggers (-> LRN high/critical)

• High-impact routines executing without human confirmation
• Lock/alarm/gas/water/heater actions under uncertain confidence

Priority Guidelines

Area Tags

Promoting to Permanent Domotics Standards

Promote only when learning is verified, repeatable, and broadly useful.

Promotion Targets

When to Promote

• Pattern recurs 3+ times within a month
• Root cause and corrective pattern are validated
• Applies across multiple homes/devices/integrations
• Includes explicit safety boundary for high-impact actuations

How to Promote

1. Distill the learning into concise, reusable rule language
2. Add it to the best promotion target
3. Update original entry:
   • **Status**: promoted
   • **Promoted**: home automation playbook (or matrix/library/safety automation)

Recurring Pattern Detection

If an event resembles existing entries:

1. Search first: rg "keyword|Pattern-Key" .learnings/
2. Add See Also links to related IDs
3. Increase priority if recurrence grows
4. Move from one-off mitigation to systemic pattern fix

Simplify & Harden Feed

Ingest strong candidates from simplify-and-harden reviews:

1. Use Pattern-Key as dedupe key
2. Search existing records: rg "Pattern-Key: <key>" .learnings/LEARNINGS.md
3. If found:
   • increment Recurrence-Count
   • update Last-Seen
   • add cross-links in See Also
4. If not found:
   • create new LRN entry with Source: simplify-and-harden

Promotion threshold:

• Recurrence-Count >= 3
• observed across 2+ areas or integration paths
• validated over at least 14 days

Periodic Review

Review .learnings/ at natural checkpoints.

When to Review

• Before major automation rollout
• After firmware updates
• After integration API changes
• Weekly reliability retrospective
• After safety-relevant incidents

Quick Status Check

rg "Status\*\*: pending" .learnings/*.md | wc -l
rg -n "Priority\*\*: critical|Priority\*\*: high" .learnings/DOMOTICS_ISSUES.md
rg -n "safety_rule_gap|device_unreachable|integration_break" .learnings/*.md
rg -n "energy_optimization|occupancy_mismatch" .learnings/LEARNINGS.md

Review Actions

• Resolve fixed entries with evidence
• Promote repeated patterns to permanent targets
• Tune thresholds and rule precedence
• Ensure confirmation steps exist for high-impact routines

Safety Posture and Operational Boundaries

This skill does not directly control physical devices. It captures and structures learnings so humans can implement changes safely.

Mandatory Safe Defaults

• Prefer notify-only mode when confidence is low
• Require explicit human confirmation for:
  • door lock or unlock routines
  • alarm arm/disarm transitions
  • gas or water shutoff actions
  • heater on/off changes with high thermal impact
• Keep conservative fallback state on sensor ambiguity
• Separate detection logic from actuator execution logic

Hook Integration

Hooks are optional and reminder-only. They must not perform direct actuator actions.

Quick Setup (Claude Code / Codex)

Create .claude/settings.json (or equivalent):

{
  "hooks": {
    "UserPromptSubmit": [{
      "matcher": "",
      "hooks": [{
        "type": "command",
        "command": "./skills/self-improving-domotics/scripts/activator.sh"
      }]
    }]
  }
}

Advanced Setup (With Output Detection)

{
  "hooks": {
    "UserPromptSubmit": [{
      "matcher": "",
      "hooks": [{
        "type": "command",
        "command": "./skills/self-improving-domotics/scripts/activator.sh"
      }]
    }],
    "PostToolUse": [{
      "matcher": "Bash",
      "hooks": [{
        "type": "command",
        "command": "./skills/self-improving-domotics/scripts/error-detector.sh"
      }]
    }]
  }
}

Enable PostToolUse only if signal detection on tool output is desired.

Available Hook Scripts

See references/hooks-setup.md for troubleshooting.

Automatic Skill Extraction

When a domotics learning is stable and reusable, extract it into a standalone skill.

Skill Extraction Criteria

Extraction Workflow

1. Identify a candidate that meets criteria
2. Run helper:

   ./skills/self-improving-domotics/scripts/extract-skill.sh skill-name --dry-run
   ./skills/self-improving-domotics/scripts/extract-skill.sh skill-name
   

3. Fill generated SKILL.md TODO placeholders
4. Update source entry:
   • **Status**: promoted_to_skill
   • **Skill-Path**: skills/skill-name
5. Validate in a fresh session

Extraction Detection Triggers

• Conversation signals: "keeps happening", "save as skill", "reusable safety pattern"
• Entry signals: multiple See Also, high/critical resolved status, repeated Pattern-Key

Multi-Agent Support

Best Practices

1. Prefer deterministic automations over hidden side effects
2. Add debounce, cooldown, and guard conditions for noisy sensors
3. Separate occupancy estimation from high-impact actuations
4. Design explicit rule precedence to avoid conflicts
5. Treat integration contracts as versioned dependencies
6. Capture before/after energy metrics for optimization claims
7. Keep manual override paths available for security and climate routines
8. Document safe fallback behavior for every critical routine

Gitignore Options

Keep learnings local:

.learnings/

Track learnings in repo: Do not add .learnings/ to .gitignore.

Hybrid:

.learnings/*.md
!.learnings/.gitkeep

Reference Links

• Skill repository: https://github.com/jose-compu/self-improving-domotics.git
• Hook setup walkthrough: references/hooks-setup.md Use this skill to improve reliability, safety, and efficiency through better documentation and review loops. Keep workflows reminder-only, and require human confirmation for high-impact routines.

Stackability Contract (Standalone + Multi-Skill)

This skill is standalone-compatible and stackable with other self-improving skills.

Namespaced Logging (recommended for 2+ skills)

• Namespace for this skill: .learnings/domotics/
• Keep current standalone behavior if you prefer flat files.
• Optional shared index for all skills: .learnings/INDEX.md

Required Metadata

Every new entry must include:

**Skill**: domotics

Hook Arbitration (when 2+ skills are enabled)

• Use one dispatcher hook as the single entrypoint.
• Dispatcher responsibilities: route by matcher, dedupe repeated events, and rate-limit reminders.
• Suggested defaults: dedupe key = event + matcher + file + 5m_window; max 1 reminder per skill every 5 minutes.

Narrow Matcher Scope (domotics)

Only trigger this skill automatically for domotics signals such as:

• sensor|actuator|automation|iot|home assistant
• energy profile|false trigger|device offline|scene failure
• explicit domotics intent in user prompt

Cross-Skill Precedence

When guidance conflicts, apply:

1. security
2. engineering
3. coding
4. ai
5. user-explicit domain skill
6. meta as tie-breaker

Ownership Rules

• This skill writes only to .learnings/domotics/ in stackable mode.
• It may read other skill folders for cross-linking, but should not rewrite their entries.