tt: Timesheet and Punch Card Verifier | Generated by AI

Home 2026.05 English 中文

---

Summary

Project: tt — Timesheet and punch card verification tool for contractor attendance compliance.

How it works:

1. User registers (tt register <id> <name>) and logs in (tt login <id>)
2. Via REPL (uv run tt with no args), user runs /add_records which:
   • Takes a screenshot of a timesheet/punch card
   • Sends it to OpenRouter vision LLM (gemini-2.5-flash) for structured extraction
   • Saves extracted data to data/<staff_id>/<record_type>.json
   • Copies the image to images/ with timestamped naming
3. tt verify cross-checks client timesheets against vendor punch cards
4. Reports weekly/monthly pass/fail against a 60% project-time threshold

Current data:

• 1 registered user: 45294575 (james.z.w.li)
• 5 weeks of client timesheet records extracted (May 2026)
• Project ratios: 80%, 80%, 80%, 100%, 100% — all passing

Bugs fixed this session:

1. uv run tt failed — missing [build-system] in pyproject.toml (hatchling added)
2. tt status crashed — .session.json picked up as contractor cache (hidden file filter added)
3. uv.lock removed from git tracking
4. GitHub remote URL updated to lzwjava/tt

Architecture notes:

• Two cache systems coexist: flat data/<name>.json (contractor verification cache from tt verify) and per-user data/<staff_id>/<type>.json (REPL records from /add_records)
• REPL has full tab completion, interactive prompts, and image extraction pipeline
• All data is Pydantic v2 models with JSON serialization

---

Back Donate