Drop the operation inventory from AGENTS.md

Merged
JE
3mo ago
+39-53

Roughly 43% of the file was a table listing all 112 operations by area. It is a
strictly worse view of spec/fizzy.smithy: the spec has the same names plus
their inputs, outputs, errors and traits, it cannot fall out of date with
itself, and it is what the pipeline actually reads. A hand-maintained copy in
an always-on file only has to drift once to start misleading. The file now
points at the spec and gives the one-liner that counts operations.

Also removed a duplicate: the upstream Fizzy source list appeared twice, once
inside the workflow steps and once as its own section. The section stays; the
workflow refers to it.

The two things worth keeping are kept -- that Fizzy has no OAuth, and the
upstream sources the spec is downstream of.

The workflow's make {lang}-generate-services was an unexpanded seed
placeholder. This repo ships all five languages but has no aggregate
generate-services target, so the per-language names are now spelled out,
including swift-generate, which does not take the -services suffix.

4844 -> 2568 bytes. Every target was checked against the Makefile at
3546f98d2cfe, and the 112 count re-derived from the spec at that SHA rather
than carried over.

Checked against main at 3546f98d2cfe5924d618c377ae07f240084578c3.

Summary by cubic

Removed the hand-maintained operation inventory from AGENTS.md and made spec/fizzy.smithy the source of truth. Tightened the workflow to avoid drift and clarified testing requirements for behavior changes.

  • Refactors
    • Dropped the 112-op table; link to spec/fizzy.smithy and suggest grep -c '^operation ' spec/fizzy.smithy for counts.
    • Clarified generation steps: after make smithy-build, regenerate all languages; code block uses make for go-generate-services, ts-generate-services, rb-generate-services, kt-generate-services, swift-generate.
    • Reinforced that openapi.json is generated and must not be edited; kept auth note (no OAuth) and MagicLinkFlow.
    • Hard Rule 3 now covers behavioral changes: require cross-language conformance tests, not just per-language unit tests.