Repository navigation
What replaces a test as the quality bar for a prose Box? #380
Description
Question
Once lib/ is gone, what stops a Box from rotting?
The old answer was a test. That answer is what built 4 431 lines of tests around 2 056 lines of code
to guard 4 095 lines of prose, and it is deleted by decisions 3 and 11. So the quality bar for a prose
Box needs naming, and the naming has to be honest about what it gives up.
The hard cases are real. test/box-staging-and-repo-pinning.test.mjs (453 lines) greps every Box YAML
for git add -A and for repository pinning; it catches a genuine class of defect, and it is also a
grep over prose, the exact antipattern. test/command-methods.test.mjs catches an upstream Method
rename that once broke /specs and /tickets with CI green.
Grill it: what replaces those two? A run that fails loudly beats a test that passes quietly — is that
enough, or does something survive? Whatever the answer, it goes in the doctrine from the §4 ticket, in
the same few lines, not in a new document.
Acceptance criteria
- The quality bar for a prose Box is stated in one short passage.
- The two hard cases above each get an explicit verdict: what replaces them, or that nothing does
and why that is acceptable. - What the bar gives up is stated plainly, not hedged.
- It lands in the root
AGENTS.mddoctrine, with no new document created.
Activity
orioltf commentedon Aug 14, 2026
How to start this session
Produces: the quality bar for a prose Box, in one short passage, appended to #374's doctrine.
/wayfinder 373 #380. It claims this ticket.test/box-staging-and-repo-pinning.test.mjs(453 lines, greps every Box YAML forgit add -Aandfor repository pinning) and
test/command-methods.test.mjs(catches an upstream Method rename thatonce broke
/specsand/ticketswith CI green)./grilling, with/codebase-designfor the deep-module vocabulary./writing-for-agentsinto the rootAGENTS.md.Traps
quietly" may be the whole answer — but say so explicitly rather than leaving it implied.
acceptable.
orioltf commentedon Aug 19, 2026
Answered in PR #398, which is also what forced the question
unic-archon-dlcwent to zero code in #381, taking twelve modules and nineteen test suites with it. Two of those suites were the hard cases named here, so the bar had to be written in the same PR that removed what it replaces.Where it lands:
#### The quality bar for a prose Box, nested inside the existing### Acceptance criteria are prosein the rootAGENTS.md. No new document and no new section at the same level (AC-4). Widening the section that already said most of it beat writing a second one beside it — two nearby passages about the same thing is the drift the one-list doctrine exists to stop. The same edit repoints that section's worked example, which citedlib/slopcheck.mjs, a file #398 deletes.The bar (AC-1) is a run and a read, in three parts: it runs where it ships (installed into a Consumer through the marketplace, no
node_modules, no hand-set environment variable, run far enough to see the step under review succeed); its rules are stated inline in every prompt that must honour them, because a rule living only in anAGENTS.mdis invisible at run time; and what it depends on is written once, in prose, with no mirror and no generator.Part one is not a preference. Every command defect of the 0.22.0 line was invisible to
pnpm testandarchon workflow listin this repository, and visible on the first command run inDXP-DesignSystem.The two hard cases (AC-2)
test/box-staging-and-repo-pinning.test.mjs— 453 lines grepping every Box YAML forgit add -Aand for a repository derived from a remote URL. Nothing replaces it. Both rules stay stated inline in every prompt that stages a path or names a repository, and both patterns are visible on the page. An unread diff now merges either one green.test/command-methods.test.mjs— the upstream-rename tripwire. Nothing replaces it either, and the reason is worth keeping: it compared two hand-written surfaces inside this repository and never watched upstream, so it could not have caught the v1.1.0 rename wave it was written for. Turning the diagnosis on the cure is what settled it. What the bar adds instead is the moment: upgrading a vendored Method Bundle means diffing the vendored tree against the new upstream tag by hand, in the commit that moves the pin.vendor/mattpocock-skills/README.mdnow says so beside the pin it holds, and the plugin version is that pin.What it gives up (AC-3)
Stated in the passage, not hedged: an unsafe
git add -Aand a repository derived from a remote URL merge green if nobody reads the diff, and between two Bundle upgrades an upstream rename surfaces on the next Consumer run rather than in CI. Both losses are real. The trade is that the alternative — keeping a module and a test whose only job is making prose greppable — is what this repo's own AC bar calls a defect rather than rigour.Closing on the merge of #398. The dependency on #381 resolves with it.