Repository navigation
Support agent-specific Smart Router Orchestrator skills - #1076
Merged
Merged
Conversation
JJ27
marked this pull request as ready for review
October 9, 2026 17:05
JJ27
added this pull request to stack #1077
October 9, 2026 17:09
JJ27
force-pushed
the
JJ27/ug-agent-specific-orchestrator-skills
branch
from
October 9, 2026 19:12
5363284 to
ae4c3d4
Compare
lilly-luo
reviewed
Oct 9, 2026
lilly-luo
approved these changes
Oct 9, 2026
GirishS2093
pushed a commit
to GirishS2093/unity-gateway
that referenced
this pull request
Oct 11, 2026
Builds on the agent-specific skill support merged in [databricks#1076](databricks#1076). ## What did you change, and why? Update the Claude and Codex Smart Router Orchestrator instructions with separate workflows for task ownership, focused evidence gathering, and independent parallel assignments. - Claude keeps connected investigations in the root when delegation would repeat the work. A delegated owner handles discovery, implementation, checks, and an integration-ready result. - Codex chooses the final-answer owner before delegation. For delegated read-only investigations that the root will present, one child acquires the complete cited evidence and the root synthesizes it. Implementation owners retain the full edit, test, and fix loop. Preserve the supplied workflows, role scopes, and execution safeguards. Keep the UG name, full activation and on/off guidance, and runtime model-evidence requirement. Adapt native Claude role names and references to the standalone plugin's unbundled configuration helper. Claude keeps its Sonnet role defaults. Codex requests Luna without forcing a reasoning effort, allowing a routed model to use its supported default. This avoids a rejected spawn when the routed model does not support `max`. One restricted fallback attempt remains for rejected bundled defaults; UG routing can replace the requested model. Explicit user and policy constraints still apply. When a harness truncates the hook output, both skills direct it to read the full agent-specific file before proceeding. Changes are limited to the two skill instruction files and their README. Both skill files use version 1.2.0. ## How do you know it works? - 33 focused tests passed across `test_skills.py` and `test_codex_routing.py`, covering installation, agent-specific hook content, and routing. - Compared each file directly with its supplied source: the original delegation rules, role scopes, coordination instructions, and integration/verification text are preserved with the agreed UG adaptations; UG's runtime-evidence guidance is appended. Codex's recovery conditions and single retry are retained, replacing only the helper-generated eligibility value with a written rule. - YAML/entrypoint checks, the standard Ruff lint/format and type-check gate, and diff checks passed. The truncated-hook recovery still needs live CI verification. Live-agent quality, cost, and latency have not been evaluated.
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?
Allow Smart Router Orchestrator to ship separate Claude Code and Codex instructions. Installation writes the selected
SKILL.<agent>.mdas the agent'sSKILL.md, and prompt/compaction hooks select the same file via an explicit agent argument. Existing skills with one sharedSKILL.mdcontinue to work.Both new files contain the current instructions unchanged. Shared role files and resources stay in place, so later instruction changes can be stacked on this PR.
How do you know it works?
src/andtests/, andty check src/, passed.Live-agent integration was not run locally. The extra documented lint check on the unchanged
scripts/run_integration.pyreports two existing TID251 violations; the normal repository lint gate passes.