Skip to content

Catch the changelog and the two build docs up - #146

Merged
davidmckayv merged 1 commit into
mainfrom
docs/catch-up-after-tonight
Aug 22, 2026
Merged

Catch the changelog and the two build docs up#146
davidmckayv merged 1 commit into
mainfrom
docs/catch-up-after-tonight

Conversation

@davidmckayv

Copy link
Copy Markdown
Contributor

Twelve merges since the last documentation pass.

Changelog

Upgrading gains the two things an operator has to know:

  • AGENT_TOOL_TOKEN is generated on a laptop now. .env.example still ships it empty and a deployment not using start.sh is unchanged, so the fail-closed default is intact.
  • start.sh no longer skips services that are already answering. A run that rebuilds an image now recreates the Bot containers, about five seconds. That is the cost of a rotated secret actually reaching the things that use it.

Fixed gains six, newest first: the Bot that browsed to a vendor it had tools for; the question routed to a coworker with no way to answer it; the deny rule walked around by computer_type with submit; the refused callback that left no trace; the request for the wheel that followed a Bot between conversations; /admin/computers listing nothing; and every component recorded twice on a first start.

Each is written as what a person running this would notice, not as what the diff did.

README: one row

Its settings table already named COMPUTER_TOKEN and SUPERVISOR_TOKEN as things start.sh sets. AGENT_TOOL_TOKEN is the third, and without it no Bot may call a tool at all — a person setting up would otherwise not learn it exists until nothing worked.

Nothing else. It says how to build and run this, not what happened to it.

docs/configuration.md

AGENT_TOOL_TOKEN was listed as unset, which stopped being true of the documented path.

It now says start.sh generates one; why a default that is right for a deployment was wrong for a laptop, where it meant every granted MCP tool was refused before it reached the grant, the boundary or the trail — and a refusal at that point does not reach the transcript, so a Bot reported no results rather than an error; and that it is one of a pair with MANAGED_AGENT_TOKEN pointing the other way, which is the part that makes rotating either of them confusing.

Also

Removes server/tests/routing.test.ts. Zero bytes — I created it by accident while looking for the real routing test and it went in with #144.

Checked, not assumed

  • .env.example already carries the corrected form-submit rule from Let the policy see the Enter a type action presses #104, and no doc repeats it, so nothing else needed changing there.
  • docs/plugins/google-drive.md is accurate as written: the redirect URI, the two APIs and the preview-program note all still hold.
  • docs/architecture.md names no audit event types, so mcp.callback_refused needed no entry.
result
bun run test 1156 pass, 8 skip, 0 fail, 1164 across 102 files
bun run typecheck clean, all four packages
bun run format:check / lint clean

The changelog gains what a person upgrading has to know and the six
defects fixed since the last pass. `AGENT_TOOL_TOKEN` is generated on a
laptop now, and `start.sh` stops skipping services that are already
answering, which costs about five seconds on a run that rebuilds an image
and is why a rotated secret now reaches the things that use it.

The README gains one row. Its settings table already named the two
secrets `start.sh` sets, and this is the third: without it no Bot may call
a tool, and a person setting up would otherwise not learn it exists.
Nothing else there changed — it says how to build and run this, not what
happened to it.

`docs/configuration.md` had `AGENT_TOOL_TOKEN` as "unset", which is no
longer true of the documented path. It now says start.sh generates one,
why the fail-closed default is right for a deployment and wrong for a
laptop, and that it is one of a pair with MANAGED_AGENT_TOKEN pointing the
other way — which is the thing that makes rotating either of them
confusing.

Also removes server/tests/routing.test.ts, an empty file I created by
accident while looking for the real one and committed in #144.
@davidmckayv
davidmckayv merged commit 2fae2fb into main Aug 22, 2026
8 checks passed
@davidmckayv
davidmckayv deleted the docs/catch-up-after-tonight branch August 22, 2026 03:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant