verisure-investigate
Structured investigation — HA logs, API trace, code trace. No code changes until approved.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Structured investigation — HA logs, API trace, code trace. No code changes until approved.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Full deploy pipeline — test, typecheck, deploy to HAOS, smoke test
End-of-session protocol — push, checkpoint, CHANGELOG, docs, todo
Session start protocol — workflow gates, pending work, status report
Bump version, publish to PyPI, create GitHub release, update HACS manifest
Dispatch 8 parallel agents for full codebase review (line-level + architecture)
| name | verisure-investigate |
| description | Structured investigation — HA logs, API trace, code trace. No code changes until approved. |
Structured diagnosis protocol. Gather evidence BEFORE proposing any fix. Do NOT write or change any code during investigation.
Launch three agents in parallel with the problem description:
Agent 1 — HA Log Analysis: Read last 100+ lines of verisure_italy logs from HA:
ssh root@homeassistant -p 22222 "docker logs homeassistant 2>&1 | grep -i verisure_italy | tail -100"
Extract ERROR/WARNING with timestamps. Identify event sequence and patterns. Also check for automation-related issues:
ssh root@homeassistant -p 22222 "docker logs homeassistant 2>&1 | grep -i 'alarm\|arming\|force' | tail -50"
Agent 2 — Entity State: Check current state of all integration entities via HA API:
source .env
curl -s "http://homeassistant:8123/api/states" -H "Authorization: Bearer $HA_TOKEN" | \
python3 -c "import sys,json; [print(f'{s[\"entity_id\"]}: {s[\"state\"]} {s[\"attributes\"]}') for s in json.load(sys.stdin) if 'verisure' in s['entity_id']]"
Compare expected vs actual. Check for stale force context, stuck states, unavailable entities.
Agent 3 — Code Tracing: Trace the code path from error/symptom. Check recent changes (git log --oneline -10). Read the relevant source files. Identify the failure point.
Correlate findings across all three sources. State root cause with evidence. If unclear, list knowns and unknowns. Propose minimal fix.
Present diagnosis to the human. Wait for approval before writing any code.
Do NOT guess. Do NOT "probably because...". Do NOT change code speculatively. Evidence first.