Skip to content

Sync .github/skills directory with azure-sdk-tools for PR 16380#10150

Open
azure-sdk-automation[bot] wants to merge 5 commits into
mainfrom
sync-.github-skills-fix/vally-skill-eval-environment-skills-16380
Open

Sync .github/skills directory with azure-sdk-tools for PR 16380#10150
azure-sdk-automation[bot] wants to merge 5 commits into
mainfrom
sync-.github-skills-fix/vally-skill-eval-environment-skills-16380

Conversation

@azure-sdk-automation

Copy link
Copy Markdown
Contributor

Sync .github/skills directory with azure-sdk-tools for PR Azure/azure-sdk-tools#16380 See eng/common workflow

helen229 and others added 5 commits July 17, 2026 18:51
Wires environment.skills: [<skill dirs>] into the mock workflow eval YAML specs so Vally loads the relevant skill per-eval instead of relying on --skill-dir at the pipeline level.
- Add missing environment.skills block to sdk-local-update-metadata-001
  (all other positive stimuli in this file already had it)
- Replace ambiguous/fictional 'contoso' package paths with the real
  fixture package (sdk/compute/azure-mgmt-compute, azure-resourcemanager-compute)
  so the model doesn't need to ask for clarification on a made-up path
- Add explicit version/release-type/date to the version-update stimuli
  since the tool genuinely requires that info

Verified locally: suite score went from ~42% to consistently 83-92%
across repeated runs (threshold 80%).
The trigger.eval.yaml suite reproducibly failed at ~70%% because vague/
informational prompts (SDK release checklist, walk me through the
release process, how to publish a new SDK version, release a new
version of azure-sdk) weren't covered by the skill's narrow trigger
phrase list, so the model didn't reliably invoke the skill.

Added those phrasings to the description, Triggers section, and
Examples, and clarified that the skill should also handle how-to/
checklist style questions, not just direct action requests.

Verified locally: trigger.eval.yaml went from 70%% to 100%% across two
repeated runs (threshold 80%%).
Per Copilot PR review feedback: anti-trigger and negative stimuli in
these eval specs didn't set environment.skills, so the skill was never
loaded/available for those prompts. That made the disallowed
skill-invocation / tool-calls grader checks trivially true (the skill
literally couldn't be invoked), providing no real signal.

Added environment.skills: [..] to all anti-trigger/negative stimuli in:
- azsdk-common-apiview-feedback-resolution/evals/trigger.eval.yaml
- azsdk-common-pipeline-fixer/evals/trigger.eval.yaml
- azsdk-common-pipeline-analysis/evals/trigger.eval.yaml
- azsdk-common-prepare-release-plan/evals/trigger.eval.yaml
- azsdk-common-sdk-release/evals/trigger.eval.yaml
- azsdk-common-generate-sdk-locally/evals/eval.yaml (sdk-local-negative-001)

Verified locally: all 6 suites still pass with the skill loaded and
available - the model correctly declines to invoke it on out-of-scope
prompts. No new failures introduced (5 trigger suites at 100%,
generate-sdk-locally-eval at 91.7%, same pre-existing unrelated failure
as before on sdk-local-analyzer-errors-001).
Copilot AI review requested due to automatic review settings July 17, 2026 19:14
@azure-sdk-automation
azure-sdk-automation Bot requested a review from a team as a code owner July 17, 2026 19:14
@azure-sdk-automation
azure-sdk-automation Bot requested a review from helen229 July 17, 2026 19:14
@azure-sdk-automation azure-sdk-automation Bot added EngSys This issue is impacting the engineering system. Central-EngSys This issue is owned by the Engineering System team. labels Jul 17, 2026
@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines:
1 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Note

Copilot could not run the full agentic suite for this review because it was automatically requested on a bot-authored pull request. Request a review from Copilot under Reviewers to retry with the full agentic suite. Improved support for bot-authored pull requests is coming soon.

Updates Azure SDK “skills” definitions and evaluation stimuli to improve routing/triggering for release workflows (release plans, generation, pipeline analysis/fixing, and APIView resolution) and to add a new pipeline-based SDK generation skill.

Changes:

  • Added per-stimulus environment.skills loading to multiple eval YAMLs to ensure the intended skill is available during evaluation.
  • Expanded trigger language and examples in release/release-plan skill docs to include more “how-to” / process-oriented prompts.
  • Added a new azsdk-common-generate-sdk-pipeline skill and updated the local-generation skill docs/evals to steer users to the pipeline flow when appropriate.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 4 comments.

Show a summary per file
File Description
.github/skills/azsdk-common-sdk-release/evals/trigger.eval.yaml Loads skill context during trigger/anti-trigger eval stimuli.
.github/skills/azsdk-common-sdk-release/SKILL.md Expands trigger wording/examples to cover release checklists/process questions.
.github/skills/azsdk-common-prepare-release-plan/evals/trigger.eval.yaml Loads skill context during trigger/anti-trigger eval stimuli.
.github/skills/azsdk-common-prepare-release-plan/SKILL.md Clarifies/expands trigger wording for release plan requests.
.github/skills/azsdk-common-pipeline-fixer/evals/trigger.eval.yaml Loads skill context during trigger/anti-trigger eval stimuli.
.github/skills/azsdk-common-pipeline-analysis/evals/trigger.eval.yaml Loads skill context during trigger/anti-trigger eval stimuli.
.github/skills/azsdk-common-generate-sdk-pipeline/SKILL.md Introduces a new skill for pipeline-based SDK generation per release plan/language.
.github/skills/azsdk-common-generate-sdk-locally/evals/eval.yaml Enriches prompts and loads skill context during eval stimuli.
.github/skills/azsdk-common-generate-sdk-locally/SKILL.md Adds guidance to route “pipeline/all languages/no local clone” requests away from local generation.
.github/skills/azsdk-common-apiview-feedback-resolution/evals/trigger.eval.yaml Loads skill context during trigger/anti-trigger eval stimuli.
.github/skills/azsdk-common-apiview-feedback-resolution/evals/eval.yaml Loads skill context during tool-call eval stimulus.

Comment thread .github/skills/azsdk-common-generate-sdk-pipeline/SKILL.md
Comment thread .github/skills/azsdk-common-generate-sdk-locally/SKILL.md
Comment thread .github/skills/azsdk-common-sdk-release/SKILL.md
Comment thread .github/skills/azsdk-common-sdk-release/evals/trigger.eval.yaml
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Central-EngSys This issue is owned by the Engineering System team. EngSys This issue is impacting the engineering system.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants