[PR #1647] [MERGED] fix(tests): separate Copilot model IDs from wire models #4724

Closed
opened 2026-10-31 00:54:23 +00:00 by bsakaguchi · 0 comments
Owner

📋 Pull Request Information

Original PR: https://github.com/morluto/rea/pull/1647
Author: @N0zoM1z0
Created: 10/10/2026
Status: ✅ Merged
Merged: 10/10/2026
Merged by: @N0zoM1z0

Base: main ← Head: fix/copilot-full-catalog-byok-alias-20261010


📝 Commits (5)

  • 8ba4717 fix(tests): separate Copilot model IDs from wire models
  • 49dd4a6 chore: merge main reference cleanup fixes
  • 0eb0db3 fix(tests): inspect parsed reference import diagnostics
  • 4cc6d0b chore: merge main setup and browser fixes
  • 641f319 chore: merge main CI fixes

📊 Changes

3 files changed (+39 additions, -4 deletions)

View changed files

📝 docs/installation.md (+7 -0)
📝 docs/testing.md (+19 -2)
📝 scripts/verify-copilot-client.mjs (+13 -2)

📄 Description

GitHub Copilot CLI can apply different client metadata to a model ID and the model name sent to a BYOK endpoint. Record and verify those separately so the native client regression can exercise a full-catalog rea-gpt-4.1 configuration while still sending gpt-4.1 on every actual HTTP request. The default configuration remains unchanged. Isolate both HOME and USERPROFILE in the disposable account.

Document the native offline BYOK recipe and its limits. The full canonical catalog is retained; the built-in gpt-4.1 ID still fails before HTTP, so #1554 remains open. The fixture requests a one-million-token prompt setting; effective client capacity and live API acceptance are unknown.

Validation:

  • Native Copilot CLI 1.0.95: full-profile alias chat and analysis, default full-profile analysis, and compact-profile built-in-ID analysis passed. Analysis loads the installed skill and reads the complete real REA JavaScript Evidence through the native spill-file tool.
  • The full-profile built-in-ID chat control still reports compaction_static_context_blocked before any HTTP request.
  • All positive cases forward the complete 170-tool catalog and validate its advertised input schemas; all verifier process lineages finish verified with no surviving descendants.
  • Final source head 641f3193247a9fa0fb4abc80c68451acb98dc5eb: the four Copilot positive cases and explicit negative control passed their expected outcomes; Qwen 0.25.0, Pi 1.1.0, Hermes ordinary/sticky profiles, and DeepSeek Harness 0.2.0-rc.2 chat/analysis also passed with complete Evidence and verified empty lineages. Qwen and Pi use native HOME skill discovery without an explicit skill setting.
  • npm run check:ci passed, including architecture guards; focused integration regressions: 38 passed, 3 skipped for Windows-only/real-browser prerequisites on this Linux host. Native Windows regression coverage remains with CI.
  • Latest main CI fixes are preserved verbatim; this PR changes only the Copilot verifier and its two guides. Exact-head GitHub CI passed all 19 checks, including native Windows and macOS package lanes.

These are installed Linux native clients with a deterministic loopback model fixture, not cloud-model or Windows/macOS acceptance. No real Hopper/Ghidra/IDA verification was required or performed for this verifier/documentation change.


🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/morluto/rea/pull/1647 **Author:** [@N0zoM1z0](https://github.com/N0zoM1z0) **Created:** 10/10/2026 **Status:** ✅ Merged **Merged:** 10/10/2026 **Merged by:** [@N0zoM1z0](https://github.com/N0zoM1z0) **Base:** `main` ← **Head:** `fix/copilot-full-catalog-byok-alias-20261010` --- ### 📝 Commits (5) - [`8ba4717`](https://github.com/morluto/rea/commit/8ba4717707c2ca49ae24813722d69bd1daac3260) fix(tests): separate Copilot model IDs from wire models - [`49dd4a6`](https://github.com/morluto/rea/commit/49dd4a695a312b5d945458bfe048eda0f5c51c06) chore: merge main reference cleanup fixes - [`0eb0db3`](https://github.com/morluto/rea/commit/0eb0db3ee1972414c03875562696a998edbde568) fix(tests): inspect parsed reference import diagnostics - [`4cc6d0b`](https://github.com/morluto/rea/commit/4cc6d0bd63e44dddd8d89590e2110bb180723e77) chore: merge main setup and browser fixes - [`641f319`](https://github.com/morluto/rea/commit/641f3193247a9fa0fb4abc80c68451acb98dc5eb) chore: merge main CI fixes ### 📊 Changes **3 files changed** (+39 additions, -4 deletions) <details> <summary>View changed files</summary> 📝 `docs/installation.md` (+7 -0) 📝 `docs/testing.md` (+19 -2) 📝 `scripts/verify-copilot-client.mjs` (+13 -2) </details> ### 📄 Description GitHub Copilot CLI can apply different client metadata to a model ID and the model name sent to a BYOK endpoint. Record and verify those separately so the native client regression can exercise a full-catalog `rea-gpt-4.1` configuration while still sending `gpt-4.1` on every actual HTTP request. The default configuration remains unchanged. Isolate both HOME and USERPROFILE in the disposable account. Document the native offline BYOK recipe and its limits. The full canonical catalog is retained; the built-in `gpt-4.1` ID still fails before HTTP, so #1554 remains open. The fixture requests a one-million-token prompt setting; effective client capacity and live API acceptance are unknown. Validation: - Native Copilot CLI 1.0.95: full-profile alias chat and analysis, default full-profile analysis, and compact-profile built-in-ID analysis passed. Analysis loads the installed skill and reads the complete real REA JavaScript Evidence through the native spill-file tool. - The full-profile built-in-ID chat control still reports `compaction_static_context_blocked` before any HTTP request. - All positive cases forward the complete 170-tool catalog and validate its advertised input schemas; all verifier process lineages finish verified with no surviving descendants. - Final source head `641f3193247a9fa0fb4abc80c68451acb98dc5eb`: the four Copilot positive cases and explicit negative control passed their expected outcomes; Qwen 0.25.0, Pi 1.1.0, Hermes ordinary/sticky profiles, and DeepSeek Harness 0.2.0-rc.2 chat/analysis also passed with complete Evidence and verified empty lineages. Qwen and Pi use native HOME skill discovery without an explicit skill setting. - `npm run check:ci` passed, including architecture guards; focused integration regressions: 38 passed, 3 skipped for Windows-only/real-browser prerequisites on this Linux host. Native Windows regression coverage remains with CI. - Latest main CI fixes are preserved verbatim; this PR changes only the Copilot verifier and its two guides. Exact-head GitHub CI passed all 19 checks, including native Windows and macOS package lanes. These are installed Linux native clients with a deterministic loopback model fixture, not cloud-model or Windows/macOS acceptance. No real Hopper/Ghidra/IDA verification was required or performed for this verifier/documentation change. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
bsakaguchi 2026-10-31 00:54:23 +00:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
morluto/rea#4724
No description provided.