Acp Issue Diagnose

gh issue view <N> --json number,title,body,author,state,labels,createdAt,comments

https://github\.com/user-attachments/files/\d+/[^\s\)]+\.(?:json|log|txt)

URL="<attachment-url>"
OUT="/tmp/acp-issue-<N>-$(basename "$URL")"
gh api "$URL" > "$OUT"     # gh handles auth; works for private repos too

curl -sSL -H "Authorization: Bearer $(gh auth token)" "$URL" -o "$OUT"

python3 -c "import json; json.load(open('$OUT'))" && echo OK

Agent(
  description: "Sonnet diagnosis of ACP diagnostics",
  subagent_type: "general-purpose",
  model: "sonnet",
  prompt: <see template below>
)

You are diagnosing an Adaptive Cover Pro diagnostics dump for a GitHub issue.

Diagnostics file: <ABSOLUTE_PATH>
Issue context (may be empty for local-mode runs):
  Number: <N or "n/a">
  Title:  <title or "n/a">
  Reporter: <login or "n/a">
  ACP version reported in body: <version or "n/a">
  HA version reported in body:  <version or "n/a">
  Cover type reported in body:  <type or "n/a">
  Reporter's description (verbatim, may be truncated):
  ---
  <first ~1500 chars of issue body, describe + reproduction sections preferred>
  ---

Your job:
1. Run the triage engine over the diagnostics file (from the repo root, with the
   dev virtualenv — the package import pulls Home Assistant):

     venv/bin/python scripts/triage_json.py <ABSOLUTE_PATH> [--latest-version <X.Y.Z>]

   That runs the SAME declarative rule table as the in-product Troubleshoot step and
   prints one line per finding plus a wiki deep link. Do NOT re-implement the checks
   by hand — the engine owns them. Pass --latest-version when the issue names the
   newest release so the STALE_VERSION check can fire.
2. Read the diagnostics file (Read tool, or Bash+Python `json.load` for large files)
   to investigate anything the engine did NOT flag. Remember the offline seam: the
   download lacks per-entity capabilities and axis requirements, so rules 8a
   (COVER_NOT_READY) and 13 (COVER_FEATURE_MISMATCH) cannot fire offline — check a
   suspected capability mismatch by hand. An unexplained symptom is a missing rule
   row: note it so the maintainer can add one (see the Developer-Triage-Rules wiki).
3. Cross-reference the reporter's described symptom against the diagnostics:
   - Does the diagnostics state corroborate the symptom? (e.g. "covers don't move" + `gave_up=true`)
   - Or contradict it? (e.g. "manual override stuck" + `manual_override_state.entries == []`)
   - Or is the symptom outside what diagnostics can show? (e.g. a UI rendering bug)
4. Identify the most likely root cause(s), ranked.
5. Note any missing information you'd need to confirm — specific entity states, HA logs around a timestamp, repro steps, etc.

Output a single JSON object with this shape — no commentary, no markdown fences:

{
  "header": {
    "integration_version": "...",
    "cover_type": "...",
    "last_update_success": true,
    "last_update_time": "..."
  },
  "critical": ["..."],            // empty array if none
  "warnings": ["..."],
  "findings": ["..."],            // bullet-form, plain English
  "decision_trace_summary": "Winning handler: X (position: Y%, reason: ...)",
  "symptom_vs_diagnostics": "corroborated" | "contradicted" | "orthogonal" | "insufficient",
  "symptom_analysis": "1-3 sentences: how the diagnostics relate to what the reporter described",
  "root_cause_ranked": [
    {"hypothesis": "...", "confidence": "high|medium|low", "evidence": "..."}
  ],
  "info_needed": ["..."],         // empty array if you have enough
  "config_suggestions": ["..."],  // concrete option changes if applicable
  "summary": "1-2 sentences for the issue reply"
}

Keep total output under ~3KB. Be specific — cite handler names, option names, entity IDs.

Looked at the diagnostics — running ACP {version}, cover type `{cover_type}`, last update {ok/FAILED} at {time}.

**What I see:**
{2–4 bullets of the most relevant findings — pull from `findings` and `decision_trace_summary`. Lead with anything in `critical`.}

**Most likely cause:**
{`root_cause_ranked[0].hypothesis` with one-sentence evidence. If confidence is low, hedge ("might be"). If multiple hypotheses are roughly tied, list the top two.}

**Suggested next step:**
{Pull from `config_suggestions` if any. Otherwise: what to try, or what info you need.}

{If `info_needed` is non-empty, add:}
**To confirm, could you share:**

- {bullet per item}

🔴 The diagnostics show {first critical item}.

I don't see a diagnostics dump on this issue yet — could you attach one?

In Home Assistant: Settings → Devices & Services → Adaptive Cover Pro → ⋮ (the row's overflow menu) → Download diagnostics. Drop the resulting `.json` file into a comment here.

If you can also include:

- A rough timestamp of when the misbehavior happened (so I can correlate against the diagnostics)
- The relevant `cover.*` entity ID

…that'd let me get to a root cause faster.