Store issue number in cluster metadata

[EivMeyer]

Problem

Heroshot currently detects which issue a cluster is working on by parsing logs:

// heroshot/src/zeroshot.ts
export function getIssueMarkerFromCluster(clusterId: string): number | null {
  const out = runZeroshot(['logs', clusterId, '-n', '200']);
  const m = out.match(/# GitHub Issue #(\d+)/);
  if (!m) return null;
  return Number(m[1]);
}

This is fragile, slow, and unnecessary - zeroshot already has the issue number, it just doesn't persist it.

Root Cause

The issue number is already available at cluster start:

// orchestrator.js:784
inputData = await provider.fetchIssue(input.issue, settings);
// inputData.number = 1157  <-- WE HAVE IT

// orchestrator.js:873 - Published to message bus
topic: 'ISSUE_OPENED',
content: { data: { issue_number: inputData.number } }

But _saveClusters() doesn't persist it to clusters.json, and listClusters() doesn't return it.

Solution

Store inputData.number in cluster state (like autoPr and modelOverride).

Changes Required (~15 lines)

1. Store issue in cluster object (orchestrator.js, in startCluster):

// After line 766
cluster.issue = inputData?.number || null;

2. Persist to clusters.json (orchestrator.js, in _saveClusters):

existingClusters[clusterId] = {
  // ... existing fields ...
  autoPr: cluster.autoPr || false,
  modelOverride: cluster.modelOverride || null,
  issue: cluster.issue || null,  // ADD THIS
};

3. Include in listClusters output (orchestrator.js):

return {
  id: cluster.id,
  state: state,
  createdAt: cluster.createdAt,
  issue: cluster.issue || null,  // ADD THIS
  agentCount: cluster.agents.length,
  messageCount: cluster.messageBus.getAll(cluster.id).length,
};

4. Load on resume (orchestrator.js, in _loadSingleCluster):

cluster.issue = clusterData.issue || null;

No New CLI Flag Needed

Originally proposed --issue <number> flag - not needed. The issue number comes from the input (covibes/covibes#1157), which is already parsed by the issue provider.

Acceptance Criteria

• cluster.issue set from inputData.number at start
• Issue number persisted to clusters.json
• zeroshot list --json includes issue field
• Issue survives cluster restart/resume
• Works for all input types (GitHub URL, org/repo#123, bare number)
• Text/file inputs have issue: null

Related

• heroshot/heroshot#3 - Blocked by this for proper cluster detection (can remove log parsing after this ships)