Skip to content

fix(configurator): guide Git setup and respect Assistant UI language - #273

Open
lweberru wants to merge 2 commits into
mainfrom
fix/assistant-guided-git
Open

lweberru wants to merge 2 commits into
mainfrom
fix/assistant-guided-git

Conversation

@lweberru

@lweberru lweberru commented Nov 18, 2026 •

Copy link
Copy Markdown
Collaborator

Closes #272.

An English UI did not pass its language to the Assistant, and a company-wide Git request led to repeated technical free-text questions. The browser now sends its selected language through chat and starting-template advice. New Git setup uses a bounded choice classifier and deterministic, source-bound recommendation policy: offer Git100/Git10 with sizing context, retain a requested company-wide standard, and present concrete access choices including deliberate reuse of saved narrow allowlists. Unknown public egress CIDRs still require the user's actual addresses.

Explicit native-field edits and existing Git configurations keep their existing guarded path. Complete proposals preserve unrelated configuration; acceptance, saving and Plan/Apply remain separate. Guidance discloses that the pinned provider replaces Git on flavor or ACL changes; this PR does not change native infrastructure or upgrade that provider.

Validation:

  • npm run check: 1,866 passed; 78 conditional tests skipped. Generated knowledge/catalogue checks, lint, typechecks and web build passed.
  • Isolated identity/database suite: 55 passed, including selected-language transport and tenant/identity preservation.
  • Desktop/mobile guided Git and discard flows: 6 passed. Four additional strict-request browser cases passed after adding the new language field to their expected payloads; the complete CI browser suite subsequently passed (558 tests). Existing English UI regression coverage was also checked. All required GitHub CI checks passed on 0925f35.
  • Real EU01 models: 10 Git cases × 2 independent iterations, 20/20 passed on final knowledge f68132e00f349065708abbe824d5c775f3010ca229a700a8789bd8bbcf830a39; suite 9cba088ee288b02aa2560b614c35e99ef59d1c9778a1d47e9cfbfe68fda270e8.
  • Complete real-model suite: 110 cases × 2 independent iterations, 220/220 passed on the same final knowledge and suite hashes. Continuous multi-topic journey: 14/14 passed, including Git setup after other services and native export; journey hash 39a6a9c881bdb47516f0f4fe381dd3c74d99ddea0945a39af1a59d9236242829. Earlier failed Git receipts are retained: they identified intent routing and saved-allowlist classification defects corrected before the successful final runs.

Synthetic tests use dedicated inference credentials and cannot call Plan/Apply. No deployment or release has been started. This improves the tested dialogue contract; it is not a guarantee that arbitrary future model responses cannot fail.

@lweberru
lweberru marked this pull request as ready for review November 18, 2026 00:35

This branch has not been deployed

No deployments
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.

Assistant: respect UI language and guide company-wide Git setup with concrete recommendations

1 participant