agentleFS
WingedGuardianWingedGuardian/skyrimvr-claude-toolkit/.agents/skills/toolkit-setup/SKILL.md

toolkit-setup

Skill21 starsChanged 11 days agoMITView on GitHub

Use when setting the toolkit up in a folder, or when a command fails because Python, Git or Node is missing, broken or too old. Inventories the runtimes by launching them, never upgrades or repairs one that exists, and installs host software only when the user's own task authorizes it.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Inventory by launching: If any Python is available, run python tools/inventory-runtimes.py inventory (here and below, python means …

    line 24
  2. Very specific:

    If no Python runs at all, do the same by hand: py -3 --version, python --version, python3 --version, git --version, node --version, and …

    line 27
  3. Very specific:

    Verify by launching again: after any install the user or you performed.

    line 40
  4. Very specific:

    Then: run bash setup.sh (Windows: from Git Bash).

    line 42
  5. Distinctive:

    Never run a package manager's upgrade, repair, reinstall or uninstall for a runtime, and never use a force, no-confirm or "latest" form.

    line 60
  6. Distinctive:

    Never download and run an installer you were not given, and never pipe a download into a shell.

    line 62
  7. Distinctive:

    Never install a runtime to make a failing check pass.

    line 63
  8. Very specific:

    Never edit the user's PATH, environment variables or registry.

    line 64

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
<!-- GENERATED by scripts/gen_agent_trees.py from agent/skills/toolkit-setup/SKILL.md. Toolkit contributors: edit the source. Your own notes belong in KNOWLEDGEBASE.local.md, which updates never replace. -->

Toolkit setup runbook

The rule

You change the user's machine only as far as the user's task says. Look first; report; wait. A missing or broken runtime is a thing to tell the user about, not a thing to fix on your own. If the task does not say, in so many words, that you may install a named runtime at an exact version, you do not install anything.

Steps

  1. Read the authorization. The task's first paragraph contains one sentence starting Host-software installs:. NOT AUTHORIZED (or no such sentence) means no installs at all. Otherwise it lists items of the form python=<winget id>==<exact version> (runtime python, git or node, then =, the id, == and the version, with no spaces; items are separated by commas). Only these winget ids are ever accepted: Python.Python.3.<minor> for python (3.10 or newer, and the version must be that minor's), Git.Git for git, and OpenJS.NodeJS or OpenJS.NodeJS.LTS for node. Any other id authorizes nothing. Anything else, a JDK, .NET or Blender included, is never covered: give the user manual steps.
  2. Inventory by launching. If any Python is available, run python tools/inventory-runtimes.py inventory (here and below, python means whichever of py -3, python or python3 starts; on Linux and macOS it is usually python3). It finds every candidate on PATH and runs each with --version only. If no Python runs at all, do the same by hand: py -3 --version, python --version, python3 --version, git --version, node --version, and look at the output, not at whether the command was found. A Windows Store alias answers with a Store message, not a version: that is not an installed Python.
  3. Plan. Save the user's whole task message, unchanged, to a temporary file outside the toolkit folder, and run python tools/inventory-runtimes.py plan --auth-file <that file>. Never pass a sentence you picked out of it: a denial anywhere in the user's text must be able to void an authorization, and the tool can only see that if it reads all of it. Delete the file afterwards. For each runtime the plan gives one action:
    • none: nothing to do.
    • report: it is there but too old, broken or hung. Do not upgrade, repair, reinstall or remove it. Tell the user what you found and the manual steps printed, then stop and wait.
    • manual: it is missing and you are not authorized. Print the manual steps for the user and wait.
    • install-authorized: it is missing, the user authorized exactly this id and version, and the command below is printed. Run only a command the plan printed, unchanged.
  4. Verify by launching again after any install the user or you performed. If the new runtime is not on PATH yet, say so and ask the user to open a new terminal; do not retry installs.
  5. Then run bash setup.sh (Windows: from Git Bash). Never pass it a flag that installs software.

House Carl is a separate optional install, not part of core setup. For an authorized integration, follow docs/housecarl.md and skill toolkit-housecarl; identify the load-order authority separately from deployment, then use its explicit setup command. Missing .NET prerequisites remain manual steps.

What an install command looks like

<!-- install-form -->
winget install --id <Id> --exact --version <V> --no-upgrade --accept-package-agreements --accept-source-agreements

<Id> and <V> come from the user's authorization, never from your memory. Only Windows with winget present is covered; on any other system the answer is always manual.

Never

  • Never run a package manager's upgrade, repair, reinstall or uninstall for a runtime, and never use a force, no-confirm or "latest" form.
  • Never download and run an installer you were not given, and never pipe a download into a shell.
  • Never install a runtime to make a failing check pass; say what failed.
  • Never edit the user's PATH, environment variables or registry.

Why Git matters

The toolkit's hooks run under Git Bash on Windows, and a hook that cannot start lets some agents run the command anyway. A missing Git is therefore a safety gap, not a convenience: tell the user.

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.