agentleFS
n2nsn2ns/hyper-hand/.agents/skills/vm-acceptance/SKILL.md

vm-acceptance

Skill5 starsChanged 24 days agoApache-2.0View on GitHub

Runs an acceptance check of a HyperHand change on the installed host and a real guest VM, following docs/manual-verification.md — preflight, temporary checkpoint, a runner script that checks independent state, evidence under build/, cleanup, and what to update afterwards (TODO, research, README). Use when a change must be verified in a real VM, or to work off an acceptance item in TODO.md.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Not passed or not run: keep or narrow the TODO item with what is still open.

    line 35

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Acceptance on a real VM

The criteria are in Manual verification: test VM baseline, independent checks, evidence and cleanup, plus setups for UAC, cold boot, enhanced session and others. This skill is the order of work. All commands run from the repository root in WSL; hh below stands for python3 client/hyperhand_client.py --vm <vm> --task-file build/<item>-<yyyymmdd>/task.

1. Preflight

  1. The automated tests pass (the windows-test skill).
  2. Install with the dev-install skill and update the VM's agent.
  3. Hashes: sha256sum "/mnt/c/Program Files/HyperHand/hyperhand.exe" "/mnt/c/Program Files/HyperHand/hyperhand-agent.exe" equal the files in build/dev-install/; hh vm_file_info paths='["%LOCALAPPDATA%\\HyperHand\\hyperhand-agent.exe"]' gives the agent's hash in the guest.
  4. hh vm_doctor: every check ok. Note what exists before the run: hh vm_checkpoints, hh vm_windows. A VM without automatic sign-in that was off waits at the sign-in screen after vm_start (agent_required); ask the user to sign in. Read the screen size from an original-size screenshot (Run.screen_size()) instead of trusting the local notes.

2. Run

  1. Write the runner as a script in build/<item>-<yyyymmdd>/ (ignored by git), so the run can be repeated. Use acceptance.py: Run(vm, item) creates that directory, its task file and the client; run.checkpoint() takes the temporary checkpoint and keeps its ID in the evidence directory (one file per VM), so a runner restarted after a crash restores the same one; run.check(name, condition, evidence) records each check and run.save() writes the results. Put the checks in try: and run.cleanup() and run.save() in finally:. Give each check its own function and run them selectively (for example through an ONLY environment variable), so a failed check can be repeated alone.
  2. Every check uses independent state, not the tool's own report: png_size of the saved screenshot, run.guest_cursor() after a click, run.process_name(pid), run.foreground(), a fixture's state file, file hashes or a fresh observation. Fixtures: run.list_fixture(counts) (Win32 lists with every item exposed), run.open_start_flyout(), and the setups in Manual verification. Stop the processes you started (run.stop_process).
  3. Change guest settings only inside the checkpoint. Host settings (enhanced session mode, a VM's memory or checkpoint type) need the user or vm_set_checkpoint_type; set them back afterwards.
  4. On a failure, first decide whether the product, the fixture or the runner is wrong; a product defect is reproduced, fixed in its own commit and the run repeated on the new build.

3. Clean up

  1. run.cleanup() restores the checkpoint and calls vm_end_turn; when the host was reinstalled during the run (a new run ID no longer owns the temp checkpoint) it removes it with all_temp: true. It checks that errors and skipped are empty, that no checkpoint added since run.checkpoint() is left (without run.checkpoint() there is no listing to compare, so take checkpoints only through it) and that a new task can write (ownership was released, with a probe that does not keep ownership itself). A failing call becomes a failed cleanup check instead of an exception. Without a runner, Run(vm, item).cleanup() restores the checkpoint of a crashed run on that VM; on a later day pass the run's date (Run(vm, item, day="yyyymmdd")), since the evidence directory is dated.
  2. hh vm_doctor is all ok and the agent's hash is unchanged; a restored checkpoint can hold an older agent, so update it again (dev-install skill, step 2) when it does.

4. Record the outcome

Results are not written up as records (document map):

  • Passed: remove the item, or its passed part, from TODO. A conclusion users rely on becomes a sentence in README's Known limitations.
  • Not passed or not run: keep or narrow the TODO item with what is still open.
  • Facts learned about Hyper-V, Windows, UI Automation or an application go to research with date, environment and how they were observed; changed behavior goes to its Features chapter.
  • Report to the user: build version, what passed, what failed or was not run, and where the evidence is.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.