Skip to content

refactor(agents): consolidate the e2e case corpus without row loss - #786

Merged
thxCode merged 1 commit into
mainfrom
case-slim
Oct 25, 2026
Merged

thxCode merged 1 commit into
mainfrom
case-slim

Conversation

@thxCode

@thxCode thxCode commented Oct 25, 2026

Copy link
Copy Markdown
Collaborator

What type of PR is this?

/kind cleanup

/area testing

What this PR does / why we need it:

The e2e skill's case corpus had grown to 122 files carrying whole families split across files that shared one fixture, one precondition and one dry-run channel, plus repeated setup blocks no helper owned. This consolidates it through a supervised four-agent pipeline:

  • Analysis (Antigravity) ranked 24 zero-row-loss case merges, 11 code extractions and 8 dead-weight cleanups, and rejected 11 merge candidates that would lose at least one STATUS row.
  • Implementation (MiniMax Code) executed the approved plan: 39 case files retired into their hosts, five new _*-lib.sh helpers carry the extracted blocks.
  • Review (Antigravity) audited the diff against the plan: 0 blockers, 2 should-fix (both fixed).
  • Adversarial crosscheck (Codex) then found four escapes the first three passes missed: a poller forwarded through a replaceable global colliding with a consumer's own signature, a teardown that stopped sweeping unnamed bindings, a node-JSON parse failure reading as an environmental skip, and a section-only Python dependency promoted to a whole-case gate. All four are fixed here.

The corpus lands at 83 cases and 13 libraries: about 3.9k lines removed (−6.5%), every logical STATUS row preserved (1,051 distinct CHECK labels before, 1,058 after - the growth is the crosscheck-era un-hiding of rows the old counter skipped and the failover suite running its labels per kill mode), and the catalogue keeps its 1:1 bijection with the disk, naming each retired number in its host row so old numbers stay recoverable.

Case numbers are not renumbered: retired numbers map to their host row, keeping historical evidence references unambiguous.

Which issue(s) this PR links to:

Relates #783

The 122-case corpus carried whole families split across files that
shared one fixture, one precondition and one dry-run channel, plus
repeated blocks no helper owned. A three-agent pipeline consolidated
it: an analysis pass (agy) ranked 24 zero-row-loss case merges, 11 code
extractions and 8 dead-weight cleanups and rejected 11 merges that
would lose rows; an implementation pass (mcode) executed the approved
plan; a review pass (agy) audited the diff against the plan; an
adversarial crosscheck (codex) then found one blocker and three
should-fix escapes - a poller forwarded through a replaceable global,
a teardown that stopped sweeping unnamed bindings, a node-JSON parse
failure reading as an environmental skip, and a section-only dependency
promoted to a whole-case gate - all fixed here.

The corpus lands at 83 cases and 13 libraries: 39 case files retired
into their hosts, about 3.9k lines removed, every logical STATUS row
preserved (1,051 distinct CHECK labels before, 1,058 after - the growth
is the crosscheck-era un-hiding of rows the old counter skipped and the
failover suite running its labels per kill mode), and the catalogue
keeps its 1:1 bijection with the disk, naming each retired number in
its host row.

Signed-off-by: thxCode <thxcode0824@gmail.com>
@gpustack-code-review

Copy link
Copy Markdown

✅ OpenCodeReview: Review skipped: no items were selected.

@thxCode
thxCode merged commit 1ceee23 into main Oct 25, 2026
9 checks passed
@thxCode
thxCode deleted the case-slim branch October 25, 2026 11:25
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