Issue Compaction

AI-powered summarization of old closed issues to reduce database bloat while preserving essential context.

Overview

Over time, closed issues accumulate lengthy descriptions, debug logs, and conversation threads that are rarely revisited. Compaction uses AI to summarize these into concise descriptions, reducing storage while keeping the important context.

How It Works

Candidates are identified — Closed issues older than a configurable threshold
AI summarizes — The description is sent to Claude (Haiku) for summarization
Original is preserved — The original description is saved in originalDescription so it can be restored
Metadata is recorded — compactedAt timestamp and compactionTier are set on the issue

Compaction Tiers

Tier	Criteria	Description
Tier 1	Closed 30+ days ago	Light summarization — key points preserved
Tier 2	Closed 90+ days ago	Aggressive summarization — essential context only

Exclusions

Issues are excluded from compaction if:

They have the no-compact label
They have no description (nothing to compact)
They are not closed
They have already been compacted

Restoring Compacted Issues

Compaction is reversible. The original description is stored and can be restored at any time:

POST /api/v1/compaction/{id}/restore

This restores the original description and clears the compaction metadata.

Background Service

A background service runs weekly and automatically compacts up to 50 candidates per run. This requires an Anthropic API key configured in your application settings:

// appsettings.json
{
  "Anthropic": {
    "ApiKey": "sk-ant-..."
  }
}

Fallback: If the AI summarization fails (e.g., API unavailable), the service falls back to simple truncation (first 500 characters + "...").

API Endpoints

GET /api/v1/compaction/preview

Preview compaction candidates grouped by tier. Returns issue IDs, titles, and description lengths.

POST /api/v1/compaction/apply

Compact specific issues by ID.

{
  "issueIds": ["abc123", "def456"],
  "tier": 1
}

POST /api/v1/compaction/{id}/restore

Restore a compacted issue's original description.

MCP Tools

Tool	Description
`analyze_compaction`	Find compaction candidates and preview what would be compacted
`compact_issues`	Compact specific issues
`restore_compacted_issue`	Restore original description of a compacted issue