webinstall/webi-installers

We're going to be creating a lot of Agent Skills because, in my testing so far, they're incredibly useful. They can help an agent use a tool, and the can also help an agent help a person use a tool. And they're more general, more useful, and faster to develop than MCP. But, the irony of https://agentskills.io/specification is that they don't yet have an agent skill for creating agent skills... so we'll need to create one. This is probably pretty easy in the sense that we can probably just give the URL and the repo to an agent and ask it to create a skill with the documentation and then do some verification that it can use it. Re #1041, #1055, agentskills/agentskills#232

How to create an Agent Skill is summarized in #661, and described in detail in https://agentskills.io/specification. We need a few skills for webi, one of which is for creating an installer. It needs to describe: How to classify a release asset (i.e. by triple host tag in the filename) How to stub out the Cheat Sheet Hot to stub out the Agent Skill How to write an asset fetcher and classifier (basically whether to use a method we already have code for - like GitHub Releases - or something specific for that tool) (this will include example code and references to existing strategies) How to test the new Agent Skill How to test the new asset fetcher and installer Re #1041

Strategies New and Updated Installers I've had an idea for a long while of adding new packages (and updating for structure changes) using heuristic-based analysis. This would obviously work perfectly for all situations in which it work, but it would take some time to build out, and it wouldn't be universal - there will always be weird edge cases. Instead, we can have Ai create canonical examples for how to list files from a tar, zip, zstd, etc use those as context to have Ai give exact commands for inspecting build artifacts, such as from GitHub Releases extract the files and get more detail from tree -AF - such as which files are actually executable have the Ai use the core examples and #412 to generate the installer and release files (and generate the code to filter out non-releases or multi-releases, such as ffprobe) Documentation Ai can be used to generate examples from a human-generated list of examples, or get ideas to supplement the material with statistically common uses or

I haven't visited this repository in about a year because every time I come here, almost everything is junk from bots or Ai (but not in a good way) or scammers or phishers or spammers and it killed my joy. I would really, really like to bring on a maintainer with triage privileges to report the junk and ping me about anything that's real, especially if it's important. I'd be happy to have other contribution as well. In particular, I think there's some ways we could use Ai that would greatly accelerate development and make it easier to get more throughput on high-quality PRs that fix problems. But I need triage help because the amount of spam and junk is so overwhelming that I lose the motivation with which I started and maintained and grew this project for years. Process Report the user. GitHub only allows you to report something like 3 issues every hour so, unfortunately, most of these junk accounts will get away without being reported because GitHub doesn't want to deal with their