# Server-only evidence for issue 23225

Issue: [#23225](https://git.cafe/pingdotgg/t3code/issues/23225).
Final head: [f88c14035e](https://git.cafe/pingdotgg/t3code/commit/f88c14035e4d4f20abbaadc16ae9d9ab40bdc52d), incorporating main [defb7e6e6b](https://git.cafe/pingdotgg/t3code/commit/defb7e6e6b). Implementation identity with reviewed head supplied by lead.

Environment: Linux x86_64, Node 24.21.0; lead-owned real isolated dev server, paired web /ws RPC, protocol 2. Cursor unauthenticated; filesystem scanning independent of account health.

## Published proof

- [Executed red log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6a8abaa63e4dd68b/red-tests.txt): regression-only execution at [23ce787387](https://git.cafe/pingdotgg/t3code/commit/23ce78738750ffa71bf8d7761e583210b452a3b4), per lead; 1 failed (missing b-small), 5 passed. Log identifies test file.
- [Final-head green log](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/faf1bc1bdfe8d1ca/green-tests.txt): 7 passed, attributed to final head by lead. Raw summary does not itself print filename, invocation or SHA. Reproduction command: `vp test run packages/provider-cursor/src/server/skills.test.ts`.
- [Base runtime failure span](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/58d0c8f5805e0b85/before-runtime.json): actual probe scan-budget-exhausted failure.
- [Final runtime transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2991ddc75c8ed3bd/runtime-transcript.md): base failure, base control, final-head success. Lead confirmed exactly 11,000 artifacts, fresh refresh RPC Success, snapshotPresent true, a-large and b-small with project scope.
- [Final runtime success span](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/10554ecff710f7c1/after-runtime.json): successful probe; inventory details are in transcript rather than this span.
- [Executable synthetic fixture recipe](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/da34960331d2e12f/create-fixture.py): create a NEW directory with two valid sibling skills and 11,000 empty files; no server launch. Stored outside repo, executable locally, not executed during capture. Equivalent ordinary-file fixture already exists in committed focused test.
- [Historical Pass 1 audit](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3b2fc75d7c7975a8/proof-audit-pass-1.md): accepts server-only proof and empty visual matrix. Earlier head, test-count and missing-evidence statements describe that earlier audit; current artifacts above supersede those evidence gaps. Its implementation observations are not a fresh review of final head.

## Scope and visual matrix

Visual rows: 0 required / 0 captured (`[]`). No rendering changes; no screenshots or recordings required. Web, desktop and mobile consume inventory downstream; no visual verification claimed. This capture step changed only external evidence files; no implementation edits, commits, pushes, fixture runs, tests, benchmarks, browser/server launches or forge writes.

## FRICTION

- Signed-in picker/real Cursor turn and macOS Electron remain unverified; accepted server-only scope does not require them.
- Lead owns runtime execution. Capture inspected/uploaded supplied artifacts; it did not independently rerun them. Final-head runtime is now supplied and published, not pending.
- Green log's exact invocation/file/SHA provenance is contextual, not embedded in its raw summary. Success span alone does not contain inventory payload; use transcript for inventory claims.
- This proves the ordinary-file regression fixture, not universal large-tree/resource bounds or a performance improvement.
- Inspected every uploaded artifact for public safety: synthetic names, temporary paths and operational trace IDs only; no credentials, pairing links or customer data found.