feat: add /backlog skill — persistent task tracking

[reshashi]

Summary

Adds a /backlog skill for persistent task tracking across Claude Code sessions using SQLite:

• 12 commands: list, add, complete, delete, search, stats, next, start, reset, get, dedup, cleanup
• Automatic deduplication — skips identical pending titles
• Priority levels: critical, important, suggestion
• Source tracking: review, qa, manual
• Integration guidance for /review and /qa — non-critical findings get captured for later

Data stored at ~/.gstack/backlog.db. Only dependency is sqlite3 (pre-installed on macOS/most Linux).

Structure

backlog/
├── SKILL.md.tmpl              # Claude instructions
└── scripts/
    └── backlog.sh             # SQLite-backed CLI

Usage

/backlog list                              # show pending tasks
/backlog add "Fix input validation" --priority important --source review
/backlog complete 42                       # mark done
/backlog search "auth"                     # full-text search
/backlog stats                             # summary by priority/source
/backlog next                              # highest-priority actionable item (JSON)

Integration with other skills

• After /review: capture non-critical findings as backlog items
• After /qa: capture bugs found but not fixed in this session
• During development: capture "noticed this could be better" moments without interrupting flow

Test plan

• backlog.sh add creates DB at ~/.gstack/backlog.db on first use
• Dedup: adding same title twice skips the second
• list --priority critical filters correctly
• search matches title and description
• next returns JSON of highest-priority pending item
• dedup and cleanup remove garbage entries
• No project-specific references in any file

🤖 Generated with Claude Code