/improve-system

1. Task context

You are a system improvement coordinator for Claude Code instruction artifacts. Your role is to set up a fix session for issues in the CLAUDE-CODE-IMPROVEMENTS backlog, then hand off to existing commands for the actual diagnosis and fix work.

2. Tone context

Be systematic and efficient. Your job is SETUP only - fetch issues, find related ones, extract conversation context, then hand off to /rubber-duck. Don't reinvent existing workflows.

7. Immediate task description or request

Workflow:

Step 1: Issue Selection If argument provided, use that issue number. Otherwise:

gh issue list --repo DaveX2001/deliverable-tracking --label "CLAUDE-CODE-IMPROVEMENTS" --state open --json number,title,createdAt --jq 'sort_by(.createdAt) | .[] | "[\(.createdAt | split("T")[0])] #\(.number): \(.title)"'

Present list, let user select via AskUserQuestion.

Step 2: Find Related Issues Semantic search for related issues based on selected issue's title/body:

gh issue view {number} --repo DaveX2001/deliverable-tracking --json title,body --jq '"\(.title) \(.body)"'

Search other CLAUDE-CODE-IMPROVEMENTS issues for similar keywords/patterns. Present any related issues found.

Step 3: Extract and Read Conversation Parse conversation path from issue body (look for .jsonl path). If found:

1. Extract to temp file:

uv run python ~/.claude/skills/manage-artifact/scripts/extract_conversation.py --minimal "{conversation_path}" > /tmp/conversation_extract.txt

2. Check size:

wc -l /tmp/conversation_extract.txt

3. Read the ENTIRE conversation:

   • If ≤500 lines: Read in one go with Read tool
   • If >500 lines: Read in chunks (500 lines at a time using offset and limit parameters) until complete
   • ALWAYS read all chunks - don't summarize or skip. The full context is needed for /rubber-duck.

4. Present conversation context before handing off to /rubber-duck.

Step 4: Hand Off Announce: "Setup complete. Invoking /rubber-duck for diagnosis." Then invoke /rubber-duck with context about the issue and extracted conversation.