The Landing climbs from the toolbox to the colleague - #9
Merged
davideimola merged 3 commits intoAug 27, 2026
Merged
Conversation
Argus derives one of two shapes from a single fact — whether an LLM provider is configured — and the Landing was written when only one of them existed. It opened at its maximum, Argus arguing back on a pull request, and then descended into configuration; the third story beat closed on "A colleague, not a toolbox" while Toolbox was a shipped shape. So the middle of the page ascends. It starts where a reader can start today, with Argus serving its scanners and what the organization knows to the AI tool they already pay for, adds the model that makes it a colleague, and only then tells the pull-request story — which now reads as what the model unlocked. Trust closes, so the boundaries land after the climb instead of interrupting it. The hero deliberately does not carry the first step: a reader at the top has not yet learned that Argus uses a model at all, so "no model required" up there answers an objection nobody has raised. ADR 0013 records the order and the exception. Toolbox and Colleague are the Guide's words, adopted rather than invented, and `agentless` is refused however well it would have fitted — in application security it already means "nothing installed on the host", and a toolbox is a daemon on the machine the code is on. ADR 0012 records both, and `CONTEXT.md` gains the three terms with the words each one is not. Two things the page now says because the Guide says them and it was about to stop short of them: the shared instance shares what Argus knows and not where it reads, since scanning stays on the host Argus runs on; and the consultation panel moved out of the toolbox section it used to share, because `consult` is served by a colleague only and a toolbox section was demonstrating the one thing that shape cannot do. The raised band was decided section by section, which survives exactly until the page is reordered. `src/landing/bands.ts` derives it from the sequence instead: the page declares its order once and every section asks, and asking about a section the order does not name throws, so a stale id fails the build rather than silently shifting the rhythm of everything under it. Three guards in `landing.test.ts` hold the rest. The refused words are read out of `CONTEXT.md`, so arming the glossary arms the test; the page has to name both shapes, which is the half that went wrong quietly while the first half passed; and no snake_case name may appear anywhere in the copy, because MCP tool names, config keys and environment variables are all snake_case and all Inventories the Guide owns.
Two things the page was getting wrong, both reported by eye rather than by a test. The threshold under "Give it a model of its own" drew a hairline across the prose, and the page draws a rule to close the hero and nowhere else — so a second one read as a stray divider instead of as a boundary. The sentence is what marks the threshold, and it does it without help. The team section was the only one arguing without evidence, on the theory that the page needed a breath there. It did not: what it needed was the argument happening, which for "the knowledge is shared and the reasoning is not" is the same window with a different mark on it, answering out of something the person asking never told anybody. So it takes the panel, on the left, and the floor's panel above it goes back to the story's default side — from there the sides alternate the whole way down with nothing repeating. Both toolbox panels now share one shape, `Exchange`, which is deliberately not a `Message[]`: on this shape the answering voice is always the reader's own client, so there is no choice left for a `from` field to record, and `asked`/`answered` say more than indexing two elements did. Which also caught a real error in the first draft of the team panel — it had Argus answering, on the one shape that cannot.
Your editor and Your AI were the two sections carrying four and five blocks of copy where every other section carries three, and it showed: the page reads as a sequence of short arguments, and two of them were essays. Your editor drops the sentence that said in eighteen words what "the whole of it: a toolbox" says in seven, and folds the scanners into the sentence that already had the knowledge in it. Your AI loses the bridge to the beats, which the threshold sentence above it was already making, and the proof that nothing is provider-specific, which the Guide is one click away for. The threshold now opens the sentence that turns instead of standing on its own. A boundary nobody crosses in the same breath is just a warning. Both sections are now the size of the ones around them: 48 and 64 words of body copy against Your team's 68, where they were 82 and 104. The beats stay lighter still, deliberately — three of them share one story and each has a panel doing the arguing.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Argus derives one of two shapes from a single fact — whether an LLM provider is configured — and the Landing was written when only one of them existed. It opened at its maximum, Argus arguing back on a pull request, and then descended into configuration; the third story beat closed on "A colleague, not a toolbox" while Toolbox was a shipped shape.
The page ascends
The spine is now: hero, the floor in the reader's own editor, the same floor across a team, the turn, the three story beats, the consultation, Trust.
It starts where a reader can start today, with Argus serving its scanners and what the organization knows to the AI tool they already pay for. It adds the model that makes it a colleague. Only then does it tell the pull-request story, which now reads as what the model unlocked. Trust closes, so the boundaries land after the climb instead of interrupting it.
The hero deliberately does not carry the first step. A reader at the top has not yet learned that Argus uses a model at all, so "no model required" up there answers an objection nobody has raised, and an opening that sells the tool the reader drives sells the product below its own ceiling. The zero-cost first step is not lost by that: it lands thirty seconds in, and in the meta description that link previews show. ADR 0013 records the order and the exception.
The words are the Guide's
Toolbox and Colleague are adopted rather than invented, and
agentlessis refused however well it would have fitted: in application security it already means "nothing installed on the host", and a toolbox is a daemon on the machine the code is on. ADR 0012 records both, andCONTEXT.mdgains the three terms along with the words each one is not.Two places the page was about to stop short of the Guide
consultis served by a colleague only. A toolbox section was demonstrating the one thing that shape cannot do.What holds it
src/landing/bands.tsderives the raised band from the section sequence instead of deciding it per section, which survives exactly until the page is reordered. The page declares its order once and every section asks; asking about a section the order does not name throws, so a stale id fails the build rather than silently shifting the rhythm of everything under it.Four guards in
landing.test.ts, each verified by mutation:CONTEXT.md, so arming the glossary arms the test;snake_casename may appear anywhere in the copy, because MCP tool names, config keys and environment variables are all snake_case and all Inventories the Guide owns;astro checkclean, 150 tests, build clean with 114 internal links resolving.