sqlprism: Skill for Claude Code

Implementing Issues

Workflow

Copy this checklist and track progress:

Implementation Progress:
- [ ] Step 1: Read issue and understand scope
- [ ] Step 2: Create branch and draft PR
- [ ] Step 3: Break down into tasks
- [ ] Step 4: Execute tasks (parallel where possible)
- [ ] Step 5: Verify and finalize

Step 1: Read Issue

gh issue view <number> --json title,body,labels,milestone

Extract from the issue body:

• Scenarios — the acceptance criteria
• Test Plan — test function names and files
• Dependencies — blocked by / blocks

If the issue is blocked by open issues, stop and inform the user.

Step 2: Create Branch and Draft PR

Follow the branch/PR naming convention: <type>-<issue_number>-<short-description>

Type mapping: enhancement → feat, bug → fix, chore → chore

git checkout main && git pull origin main
git checkout -b <type>-<number>-<description>
git push -u origin <type>-<number>-<description>

Create a draft PR immediately so progress is visible:

gh pr create --draft \
  --title "<type>-<number>-<description>" \
  --body "$(cat <<'EOF'
Closes #<number>

## Summary
- <to be updated on completion>

## Test Plan
<copy Test Plan table from issue, add Status column set to pending>
EOF
)"

Step 3: Break Down into Tasks

Analyze the issue scenarios and create a task breakdown. Save to .claude/plans/<issue_number>/tasks.json.

mkdir -p .claude/plans/<issue_number>

See TASK-FORMAT.md for the JSON schema and decomposition rules.

Key rules:

• Each task maps to one or more scenarios from the issue
• Tasks that touch different files/modules can run in parallel
• Tests are a separate task per module, but run after implementation tasks
• Every task has a clear "done when" that maps back to a scenario

Step 4: Execute Tasks

Read .claude/plans/<issue_number>/tasks.json and identify which tasks can run in parallel (no shared files, no dependency between them).