PATAPIM Deep Test
Full-app test sheet β 982 items across 20 sections
(plan 1.0.1). No account needed: enter your name, start a run, check things off.
Multiple people (and Claude agents) can run sheets at the same time.
How testing works
π§ You test by hand
- Enter your name below, pick the environment you're testing on, hit Start run.
- The sheet opens: work through it section by section. Each item tells you the preconditions, the steps to perform in PATAPIM, and what's expected (click βΈ on an item to see them).
- Mark each item: β pass β fail β blocked βΈ skip β and write a note whenever something is off. Everything autosaves; you can close the tab and resume from the runs list anytime.
- When you're done (or done enough), click Finish & export report β it files a markdown report in the archive.
π€ Claude tests for you
- Open Claude Code in a normal terminal, outside PATAPIM, on the machine where PATAPIM is running (host or VM).
- Make sure Computer Control is ON in PATAPIM Preferences, and the
patapim-browserMCP server is added (the kickoff prompt below explains how). - Paste the kickoff prompt below. Claude claims its own run, executes every agent-testable item via MCP, posts results here as it goes, and files its own report. Expect this to take hours β that's the point.
- Watch its progress live from the runs list (refresh; results can lag ~1 min).
π» Full VM pass (the real deal)
- Create a fresh Windows 11 VM (Hyper-V Quick Create works). Snapshot it before installing anything.
- Inside the VM: download PATAPIM from patapim.ai/download and install. Start a run named like
VM-firstrun-<date>and do the Onboarding & First Run section FIRST β it's the only environment where onboarding is testable. Restore the snapshot to replay it. - Then continue by hand and/or install Claude Code in the VM and run the agent flow there too.
1 Β· Start a new run
2 Β· Runs β resume yours or watch others
Loadingβ¦
3 Β· Claude kickoff prompt
Copy this into a Claude Code session running outside PATAPIM on the test machine. It contains everything Claude needs.
Full protocol & pitfalls: runbook.md. Raw API, if you're scripting it yourself:
# 1. claim a run
curl -X POST https://patapim.ai/api/deep-test/runs \
-H "Content-Type: application/json" \
-d '{"name":"Claude","kind":"agent","environment":"host stable"}'
# 2. fetch the plan (filter items: mode in ["agent","both"])
curl https://patapim.ai/api/deep-test/plan
# 3. post results incrementally (batches β€25, β₯5s apart)
curl -X POST https://patapim.ai/api/deep-test/runs/<runId>/results \
-H "Content-Type: application/json" \
-d '{"by":"Claude","results":[{"itemId":"terminals-001","status":"pass","note":"β¦"}]}'
# 4. file the final report
curl -X POST https://patapim.ai/api/deep-test/runs/<runId>/report \
-H "Content-Type: application/json" \
-d '{"markdown":"# Report β¦","by":"Claude"}'