Selora MCP Skill

Purpose

Use Selora MCP tools to move from read-only context to validated and explicitly approved home automation changes.

Preconditions

1. Selora MCP server is reachable at /api/selora_ai/mcp.
2. A valid Home Assistant Bearer token is configured.
3. For admin-gated or mutating tools, caller has Home Assistant admin privileges.

If preconditions fail, stop and report what is missing.

Trigger Phrases

Activate this skill when requests include language such as:

• "analyze my home"
• "show/list my Selora automations"
• "validate this automation YAML"
• "create/enable/delete this automation"
• "list suggestions" or "show proactive suggestions"
• "list patterns" or "show detected patterns"
• "accept/dismiss this suggestion"
• "run/trigger a Selora scan"
• "continue my Selora chat/session"

Do Not Trigger For

Do not activate this skill when the request is primarily about:

• Generic Home Assistant troubleshooting unrelated to Selora MCP tools
• Frontend/UI styling, dashboard layout, or Lovelace card design changes
• Non-Selora automation management unless explicitly requested to migrate into Selora flow
• Pure documentation writing that does not require MCP tool execution

Tool Surface

Read/context tools:

• selora_get_home_snapshot
• selora_list_automations
• selora_get_automation
• selora_validate_automation
• selora_list_sessions
• selora_chat (admin-gated, stateful)
• selora_list_patterns
• selora_get_pattern
• selora_list_suggestions

Mutating/admin-gated tools:

• selora_create_automation
• selora_accept_automation
• selora_delete_automation
• selora_accept_suggestion
• selora_dismiss_suggestion
• selora_trigger_scan

Identifier Integrity Rules

1. Never invent automation_id, session_id, pattern_id, or suggestion_id values.
2. Resolve IDs from tool output only.
3. If target identity is ambiguous, ask for clarification before mutating calls.

Safe Call Sequencing

A) Home context and discovery

1. Call selora_get_home_snapshot first.
2. Use selora_list_automations / selora_get_automation for automation context.

B) YAML-first automation flow

1. Call selora_validate_automation with external YAML.
2. If invalid, return errors and stop.
3. If valid, show normalized YAML + risk.
4. Ask explicit confirmation.
5. Call selora_create_automation with enabled=false unless user explicitly requests immediate enablement.
6. Optionally call selora_accept_automation after explicit approval.

C) Natural-language automation flow

1. Call selora_chat with message (and session_id if continuing).
2. If automation YAML is returned, summarize risk and ask for explicit approval.
3. Only then call selora_create_automation or selora_accept_automation.

D) Pattern/suggestion flow

1. Optionally call selora_trigger_scan when user asks for fresh results.
2. Call selora_list_suggestions and/or selora_list_patterns.
3. If needed, call selora_get_pattern for full detail.
4. Ask explicit confirmation before selora_accept_suggestion or selora_dismiss_suggestion.

Admin and Write Boundaries

Rules:

1. Never execute mutating tools without clear user authorization in the current thread.
2. For create/accept operations, default to disabled-by-default unless explicitly requested otherwise.
3. For delete/dismiss operations, require explicit confirmation.
4. If auth/admin checks fail, stop and report.

Confirmation Protocol

Use explicit confirmations such as:

• Yes, create automation <automation_id or alias>.
• Yes, enable automation <automation_id>.
• Yes, delete automation <automation_id>.
• Yes, accept suggestion <suggestion_id>.
• Yes, dismiss suggestion <suggestion_id>.

If intent is implied but not explicit, ask and wait.

Risk-Gated Policy

1. Always surface risk_assessment before mutating automation or suggestion actions when available.
2. If risk is high or safety flags are present, require a second confirmation.
3. If risk data is missing, state that and require confirmation.

Don't Do This

1. Do not skip validation for externally provided YAML.
2. Do not invent IDs.
3. Do not bulk-mutate without explicit user request.
4. Do not silently enable automations after create.
5. Do not mutate non-Selora automations through Selora lifecycle tools.

Output Contract for Agents

After tool calls:

1. State what was read or changed.
2. Include key IDs.
3. Include risk summary for any proposed or accepted automation.
4. Before mutating calls, provide a concise confirmation prompt.

Platform Notes

• Compatible with Claude Code, Codex, OpenClaw, Cursor, and Windsurf.
• If runtime supports allowed-tools, align it with this skill’s read-first and confirmation-gated mutation policy.

Synchronization Source of Truth

Canonical source file:

• .claude/skills/selora-mcp/SKILL.md

Mirror targets that must remain identical:

• .codex/skills/selora-mcp/SKILL.md
• .openclaw/skills/selora-mcp/SKILL.md
• .cursor/skills/selora-mcp/SKILL.md
• .windsurf/skills/selora-mcp/SKILL.md