bb-plugin-bus
AGENTS.md0 starsChanged 2 months agoMITView on GitHub
A typed peer bus between bb threads: twelve closed message kinds, TTL'd resource claims, delivery on bb's own send primitive. README.md is the user-facing document and the design rationale; skills/bus/SKILL.md is what agents are taught. A behaviour change owes an edit to the skill too — the live suite asserts that skill's kind table equals KIND_NAMES, so the debt is checked rather than remembered. npm test, npm run typecheck, and two GitHub Actions workflows — test.yml and managed-install.yml. The second…
- Installs packages
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 50
Nothing deployment-specific in this public repo: no host, server or person names as defaults in code or README.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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.
Files
<!-- agents-md ceiling: 60 lines -->AGENTS.md — bb-plugin-bus
A typed peer bus between bb threads: twelve closed message kinds, TTL'd resource claims,
delivery on bb's own send primitive. README.md is the user-facing document
and the design rationale; skills/bus/SKILL.md is what agents are taught. A behaviour
change owes an edit to the skill too — the live suite asserts that skill's kind table
equals KIND_NAMES, so the debt is checked rather than remembered.
Commands, all run 2026-09-09
npm install # rc=0
npm test # node --test over src/*.test.ts — 151 tests, 0 fail (2026-10-04)
npm run test:live # six cases against a REAL bb and two spawned threads
npm run typecheck # tsc --noEmit, rc=0
bb plugin build . # dist/server.js + dist/server.meta.jsonThe gate
npm test, npm run typecheck, and two GitHub Actions workflows — test.yml and
managed-install.yml. The second is the one a clean local run cannot stand in for: bb's
managed git install resolves runtime dependencies only (--omit=dev --omit=optional
--ignore-scripts) before bb plugin build, so a module imported at runtime but parked in
devDependencies builds here and fails for every real user. The dependencies block is
empty and that is a feature; adding the first one is a decision, not a detail.
npm run test:live is not in CI and cannot be: it needs a running bb and spawns real
threads. It SKIPS when bb is down, and a skip there is a finding, not a pass.
Layout
The unit-by-unit table is in README.md — one home, and it is
the one a contributor reads first. The single fact that governs edits here:
src/kinds.ts is the kind table, and everything else derives from it.
Conventions that differ from the defaults
- Tests are
node --test --experimental-strip-typesagainst the sources, no build step, no vitest.npm testnames each file literally so the live suite cannot be swept in, andsrc/suite.test.tsasserts that list against the directory — a suite file nothing runs is a test nobody has. There is no second list of kinds: adding one is a row insrc/kinds.tsand a test. - Nothing here waits, blocks or polls — for the LOCAL bus. Delivery is the bb server's job; a listener, a poller or a queue drain breaks the property the design rests on — that a session cannot silently go deaf. Two exceptions: the wait list's single expiry timer (MX-1390), and the shared space's sync service (MX-1528), which exists because nothing may connect into a server, does nothing until configured, and logs every failure.
- Nothing deployment-specific in this public repo: no host, server or person names as defaults in code or README. The shared space takes all of that from settings.
- Two facts about bb that constrain every change here — a body can only come from
--body-file(bb forwards neither argv-safe prose nor stdin to a plugin CLI), andqueuestill wakes an idle thread. Both are measured, with their incidents, inREADME.md; do not re-derive either from the code.
Nothing about who may merge, how agents are spawned, or how the maintainer's machine handles secrets belongs in this file, and none of it is stated here. Those are properties of a working environment, not of this project; if you are contributing, your own conventions apply and nothing here depends on the maintainer's.
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
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.

