Solo Dev AI Team Ops

The problem

You give your agent three fixes and say "start with the first one". It fixes one and stops. Tomorrow’s session reads a handoff that still says the bug is broken — if it reads it at all. After /clear, CLAUDE.md and auto memory come back, but not where the work stands: which task is next, what changed since the notes were written, whether the notes are still true.

The four paid skills

• overnight-resume: a checkpointed task queue (title | check command). A task is done only when its check passes; 3 failures = blocked and skipped; "needs a human" = parked. State is saved after every step, so after a crash, rate limit or new session the agent picks up after the last finished task. Morning report included.
• preship-ui-check: screens × window sizes × {normal, empty, bad input, wrong order}. Refuses SHIP while a cell is blank, a bug is open or the date/build header is empty. Dead buttons count as bugs.
• cross-review: a review packet for a read-only second reviewer (Codex, Gemini, a fresh session, a human). Common key/token/e-mail formats are redacted (best effort — read it before sending); refuses when .env or key files are in the diff. The skill tells the agent to reproduce each finding before fixing.
• i18n-qa: missing keys, broken placeholders ({name}, %s, <b> and more), untranslated and too-wide CJK strings. JSON (i18next etc.), Chrome _locales, Apple .strings, Android strings.xml, Godot/Unity CSV.

Also included

• team-ops: one init per project adds a Claude Code SessionStart hook (on a new session, /clear and auto-compact the agent gets your handoff and the next task) and a Stop hook (sends it to the next open task, and reminds it to refresh the handoff when commits get ahead of it). Your settings.json and CLAUDE.md are merged and backed up, never replaced; remove undoes it.
• handoff-file (free, MIT): a handoff draft prefilled from git, plus a checker. The free skill is included in the download.

Recorded with a real session

A real Claude Code session (Haiku, sample project with 3 tasks), told "Start on tasks.txt and give me a short report after the first task": without the pack, 1 of 3 tasks and the handoff unchanged (4 of 4 runs). With the pack, 3 of 3 tasks and the handoff refreshed and checked (4 of 4 runs).

We also recorded where it made no difference: in a 5-file project, when you just say "continue", the agent finds the notes by itself. The full record, a script to re-run it with your own Claude Code, 47 automated tests and a before/after demo are in the download.

Works with

Claude Code. The hooks are Claude Code only; in Codex and Cursor the skills and scripts work the same, and teamops.py brief starts a session with the same information. Python 3 only — no packages, no accounts, no API keys. The scripts make no network calls themselves (your own check commands run as you wrote them). Read the scripts before you run them.

Honest limits

• It does not restart a crashed or closed session; a new session resumes from the saved queue.
• It does not make the agent smarter. Good check commands are what make "done" trustworthy.
• Redaction catches common formats, not every secret. The UI matrix does not click buttons for you.

Price

US$19 one-time · 4 paid skills + team-ops + free handoff-file · 12 months of updates.

License

Use it in your own personal and commercial projects, on any number of your own computers. One purchase = one person. No resale, redistribution or publishing (including GitHub or skill marketplaces), in original or modified form. You may modify the files for your own use. (The free handoff-file is MIT.)

Refunds

If Solo Dev AI Team Ops does not work for you, email [email protected] within 14 days of purchase and we refund the full price. No questions asked. Refunds go back to the original payment method through Creem and usually appear in 5–10 days.

Purchase