Repository navigation
Conversation
JJ27
force-pushed
the
JJ27/ug-orchestrator-ci-verification
branch
from
October 8, 2026 01:26
ae0f4fa to
14b8ba0
Compare
rohita5l
pushed a commit
to djcelis/unity-gateway
that referenced
this pull request
Oct 8, 2026
## What did you change, and why? Add Smart Router Orchestrator to existing Claude and Codex smart-routing launches behind the default-off `ENABLE_SMART_ROUTER_ORCHESTRATOR=1` flag. Opted-in launches install `smart-router-orchestrator` alongside `smart-router`, load the bundled Claude roles in the existing temporary plugin, and register hooks before root prompts and after compaction. The hooks use the existing UG session and routing controls to supply the workflow or tell the root that orchestration is off. Turning Smart Router off through its skill stops new automatic delegation; turning it on restores orchestration only in opted-in sessions. Changes made outside the conversation are observed at the next prompt or compaction hook. Explicit user requests for subagents remain supported while routing is off. Smart Router selects child models; the skill supplies role instructions. Claude roles retain their Sonnet defaults, and Codex delegates without model or reasoning-effort overrides. Codex combines UG's launch hooks with native user/project hook sources and applies project trust. Existing launch eligibility and session setup are reused. Update the skill instructions and documentation for this behavior. Extend the real Claude/Codex toggle journeys to cover the orchestration flag off/on and verify installed skills, saved controls, actual command results, completed child tasks, and correlated routing decisions. ## How do you know it works? - 176 local tests passed across skill installation, Smart Router controls, Claude/Codex launches, integration contracts, and evidence readers. - Built the wheel offline and checked the packaged `smart-router-orchestrator` skill in temporary Claude and Codex homes. Both prompt and compaction hooks returned the full skill with `ENABLE_SMART_ROUTER_ORCHESTRATOR=1`; flag-off behavior and bundled Claude roles passed. - Repository Ruff lint, formatting, type checks, and diff checks passed. The toggle journeys explicitly request children. Automatic delegation and compaction are tested separately in databricks#1036. Native-agent results for this revision are pending CI; the local bundle checks do not establish a live agent pass. --------- Co-authored-by: lilly.luo <lilly.luo@databricks.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What did you change, and why?
Keep temporary live verification separate from the integration changes in #996. This draft targets that PR's branch and changes only tests and their documentation.
Two real-agent journeys exercise Claude and Codex with
ENABLE_SMART_ROUTER_ORCHESTRATOR=1. Each requests a review of three source modules without asking for subagents, requires completed native child work and a root report containing values read from all three files, performs native/compact, and repeats with different files. Both reviews require fresh workflow context. Codex delivers its compact hook before the next model request, so the follow-up review checks that delivery.The journeys use the existing managed-config test lanes and real gateway and agent binaries. They check delegation and continuation, not the accuracy of review findings. There are no production-code or shared CUJ-fixture changes.
How do you know it works?
d4a505c. Codex hit model API 429 rate limits in both attempts, so its complete journey remains unverified. Results for the renamed revision are pending CI.This is a temporary validation PR, not intended for merge.