refactor: Import HAL repo

.github/typos.toml

@@ -1,3 +1,6 @@
 
 [files]
-extend-exclude = []
+extend-exclude = [
+  "_import-HAL",
+  "_import-artifact"
+]

_import-HAL/README.md

@@ -0,0 +1,165 @@
+Isolates: pieces of conventional code packaged in such a way as to be callable
+by AI models. These code calls can include API calls to external services and
+other side effect inducing things.
+
+Helps: pieces of text that contain instructions for an agent to expand and fine
+tune their abilities to get jobs done, instead of packing everything into the
+system prompt. The instructions might involve calling on other agents
+
+Agent: consists of a system prompt, optionally isolates, a filesystem, and
+helps. When they are invoked using an AI model, the invocation becomes a Thread.
+
+Threads: A running agent, containing stateful conversation and the filesystem in
+a possibly mutated state. The index of threads is available to Backchat.
+
+Backchat: A special thread aware thread designed for navigation thru all the
+threads, branches, commits, and files.
+
+Actor: A branch representing an identity on the system. This is commonly a
+human, but can also be a machine. It represents a permission domain. It contains
+one or more machines.
+
+HAL: the base agent that everyone starts with
+
+# Helps
+
+If AI is a universal function that represents a summation of every function ever
+written, then helps are the specific parameters with which to call that function
+to get a reliable result back
+
+Helps are required to be in machine readable format since loading them up and
+then passing thru an AI just to get out the instructions seems frivolous.
+
+We can present them for being edited using human friendly mechanisms, but they
+do need to be loaded by machine ultimately, so it saves time to store them in
+this machine readable fashion.
+
+They represent a set of instructions for how to be called by an AI, including
+the tools that that AI needs.
+
+The help runner is a standard piece of code that knows how to execute helps in
+this format, making it possible to supply it with a different help and get
+wildly different behaviours out of it. It can support a range of different
+models that can be drawn upon, and can have any range of tooling supplied to
+make conventional function calls.
+
+Helps can be nested so that they can reply on other helps, like dependencies.
+
+It is how to call something, not how to get something that will be called a
+certain way - it is function invocation. It is a collection of function
+invocation parameters, tested to have reliable outputs, indexed by the problems
+they are trying to solve, to make it more accessible to an AI trying to solve
+problems on the fly. This is different to npm, where npm is a collection of
+things to be invoked, not the invocations themselves, altho npx acts in the same
+way.
+
+The commands are isolates - pieces of code that follow a standard format for
+running in a git based system. The runner is an isolate too, but it is distinct
+in that it is used to boot from ?
+
+The ability to call another help file is provided to each help file optionally.
+
+They're almost like bottled function parameter calls that have some known good
+effect, where the function always takes just a single text parameter ?
+
+Isolates have code, help have function instructions to call that code.
+
+Helps are the interface between human / AI readable and machine code readable.
+
+Think helps are only ever invoked with a single parameter - the text prompt.
+They are natural language interpreted functions.
+
+The return for calling a help as a function should have a standard format.
+status: DONE | PENDING | ERROR | NOCANDO output: the NL output of the function -
+there seems no point using json or any other format. To output a given format we
+would need to have called a tool that outputs that format. NOCANDO means that
+what was asked for is outside the competency envelope of this help. PENDING
+means it needs more information to continue.
+
+Might consider returning a confidence rating with results.
+
+Runners are a subset of isolates, and are called by isolates, so have access to
+the io hooks that isolates do. They have a different interface and specialize in
+the operation of help files, where the help is dynamic.
+
+## Sections
+
+All sections are optional, and with nothing specified, a deprompted AI will be
+called with the default model.
+
+### Config
+
+This is a freeform area that can be used to store any data needed for the
+execution of the help. We might move this section to be inside of Runner
+
+### Runner
+
+Different AI models need different runners. The type of runner needs to be given
+with the help so we know what to call. In some cases we may need to go fetch
+more code to be able to run these other AI models in an isolated environment.
+
+It says that this piece of code is supposed to be run using this Help, which
+includes this set of config parameters
+
+### Commands
+
+This the list of function names and paths to access the commands that will be
+run. These will be loaded up to be available to the AI to call as it wishes. The
+API defined at this location will inform the model of the parameter format and
+purpose.
+
+### Instructions
+
+This is the system prompt given to the AI that will run this help. Paragraphing
+is done by writing as array entries, which are all joined with a newline
+
+### Done
+
+This is a check for when a complex operation has been done.
+
+### Examples
+
+Helps with making this help retrievable in goal space, but also
+
+### Tests
+
+A test suite can be either referenced or written here, so that when the help is
+being tested, it will be exercised by the contents of this section. These would
+take the form of an input text prompt and some expected output, where the
+overall Done condition is also evaluated.
+
+## The Runner conventions
+
+If make a reply convention where if the runner needs more info, it invokes the
+moreInfo() function and then it gets passed back in to the caller, which then
+figures out what it needs and then makes another call back. This call back is
+crossed over as the response to the original call, so the original call can then
+be resumed with the new information.
+
+We can have similar conventions for progress updates being sent back for long
+running processes.
+
+The chat session is preserved by way of it being dedicated and in its own
+branch. This might mean that actions should be prefixed by the branch that
+called them ? or can they be blanked upon merge, since the history can be found
+by tracing the files ? Tracing branches means the actions would be already
+filtered by the process that was calling them.
+
+## Examples
+
+### Using Help as the user prompt
+
+When the system first boots, we have to support a prompt response system. This
+is provided under the hood by a help. This first help is the goalie, and it will
+loop around using the 'stuck-finder' function until it has a help that it is
+happy about using. Once it is done, it will call the runner with the path to the
+help that it found.
+
+At the start, the stuck finder function is just an AI function
+
+### Using Help as pure executable
+
+We should be able to specify pure code that runs using the help format. This
+would be no different from an isolate. Some commands, in fact, will be helps at
+the end of IO queues that can called on directly to perform tasks, rather than
+dynamically like helps do - ie: they are permanent functions, not jitters.

_import-HAL/agents/agent-fixture.md

@@ -0,0 +1,9 @@
+---
+commands:
+  - io-fixture:local
+  - io-fixture:error
+---
+
+Your name is 'Assistant'
+
+ALWAYS be as brief as possible

_import-HAL/agents/cheeky-bstrd.md

@@ -0,0 +1,13 @@
+---
+description: A specific bot to deal with only questions on the pre-specified list below.
+config:
+  model: gpt-4o
+---
+
+You are a humourous bot that CAN ONLY answer questions on three topics.  If the user asks any question on any other topic, you are to say "Duh?  How would I know?"
+
+Here are the three topics which you can answer:
+
+1. Questions concerning the fit of clothing for women.  You are to be polite, and act like a gay man who is her best friend.
+2. Questions concerning politics.  You are to be a comedian when replying.
+3. Questions concerning lamb rogan josh.  You are always to respond with something like "Jude makes a great lamb rogan josh.  Why are you asking me?  Ask her!"

_import-HAL/agents/create-agent.md

@@ -0,0 +1,5 @@
+Give it the format of a blank file. Teach it how to write these files out.Once
+written, it is automatically picked up by the help-finder. We should guide the
+user thru how to run some tests, and show the default tests that are
+automatically running on it, with the examples being used to test its outputs
+and expectations.