Repository navigation
Add agentic issue investigation workflow - #50606
Open
ROHIT SINGHAL (rohitsinghal4u) wants to merge 10 commits into
Open
ROHIT SINGHAL (rohitsinghal4u) wants to merge 10 commits into
ROHIT SINGHAL (rohitsinghal4u) wants to merge 10 commits into
Conversation
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Azure Pipelines: Successfully started running 1 pipeline(s). 34 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run. |
Contributor
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
Unresolved workflow security and correctness issues affect command permissions, checkout isolation, failure reporting, and triage output execution.
Review effort: Lite
Findings: 1
Open (2)
What changed in this PR
Adds a post-triage agentic issue investigation workflow for Azure SDK for Java.
Changes:
- Adds investigation rules, decisions, and safe outputs.
- Integrates triage handoff and owner-recovery ordering.
- Adds Key Vault troubleshooting context and generated workflow locks.
| File | Summary |
|---|---|
sdk/keyvault/known-behaviors.md |
Adds advisory Key Vault service context. |
.github/workflows/issue-triage.md |
Adds investigation handoff logic. |
.github/workflows/issue-triage.lock.yml |
Regenerates the triage workflow. |
.github/workflows/issue-investigation.md |
Defines investigation behavior and outputs. |
.github/workflows/issue-investigation.lock.yml |
Compiles the investigation workflow. |
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Use the native close_issue body for the complete service-side explanation, instead of posting a separate comment before closure. Add source and generated-output regression checks to the existing Node test step. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Keep file citations tied to the repository, ref, and path successfully read. Verify each citation before publishing rather than replacing fork evidence with an unverified upstream-main URL. Retain trusted external metadata sources and add a guidance regression contract. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Restore the existing triage source, lock and actionlint workflow exactly. Dispatch investigation from a separate completed-run handoff with bounded artifact provenance, native threat parsing and fresh issue gates. Keep regression validation separate from existing jobs and cover owner failure recovery and handoff corner cases. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Use the pinned producer whitespace semantics rather than Python strip defaults and reject non-finite JSON evidence. Compare body hashes directly with the pinned runtime implementation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…te evidence Add an independent read-only preflight that emits a local noop for invalid issue numbers before native AI/reporting jobs can run. Do not treat shared missing diagnostics or earlier unverified automated assertions as proof of duplication. Preserve native detection for valid requests and all legacy workflow definitions. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Apply current evidence rules before comment deduplication and distinguish lagging catalog entries from verified latest publication evidence. Preserve valid unchanged outcomes and scope reproduction requests to older reported versions. Add regression contracts without changing runtime or legacy workflows. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This branch has not been deployed
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.


Description
Adds generalized post-triage issue investigation for Azure SDK for Java, based on Azure/azure-sdk-for-net#60514, without modifying the existing triage workflow.
Existing workflow compatibility
The initial integration incorrectly made regular triage outputs depend on owner notification. The implementation now restores
issue-triage.md, its generated lock andactionlint.ymlexactly to their base/main definitions.A separate completed-
workflow_runhandoff checks investigation eligibility. It cannot delay or suppress the completed triage run's labels, assignments, comments, recovery or completion result. New tests run in a separate validation workflow.Both live automatic legacy paths passed: prelabelled exit and manual-triage fallback. Their separate followers started after completion and suppressed investigation dispatch. The original fixture25 event was dropped while fork Actions were disabled; an already-approved prelabelled fixture26 supplied the fresh automatic probe. The dropped event remains preserved as an infrastructure observation.
Investigation and handoff
Reassessment and metadata clarification
The latest live follow-up identified a conflict between duplicate reassessment and comment deduplication. Preview31 still nooped by accepting its old unsupported automated duplicate assertion.
The new guidance makes reassessment happen first. An unsupported automated duplicate outcome is corrected once with a targeted information request; a supported existing outcome and equivalent next action still produce noop on unchanged reruns.
Management34 correctly requested missing information, rejected the unsupported duplicate and did not request a downgrade. Its remaining strict-test failure was unclear treatment of catalog2.55.5 versus newer Maven2.55.6. Catalog entries are now explicitly source-attributed candidates, not proof that no newer publication exists. Conflicting publisher evidence must be reconciled or reported as unresolved, rather than calling the older entry a verified latest version. The older-version reproduction rule is explicitly scoped to older versions.
The metadata implementation, schema, credentials, network destinations and version ordering policy are not changed by these wording corrections.
Validation and live results
b6e8062a3a, all13 targeted runs executed:11 passed and2 failed, with no infrastructure or dependency blocks. All five malformed inputs ran only preflight and made zero repository mutations. Valid-input customer/PR rejection, service closure, exact duplicate and unchanged reruns passed.not_planned; its rerun made no changes. Precise duplicate49 linked issue29 with one comment; its rerun made no changes.b6e8062a3a, including macOS and Azure build6971110, passed. The older macOS download failure was superseded by that successful build; no manual retry succeeded.Scope and limitations
No Java SDK API, dependency, release-version or existing-workflow changes. No new secrets, external LLM endpoints or runtime upgrades.
The pinned runtime has no independent detection-tracker opt-out. Malformed inputs now cannot reach that reporter, but warnings on valid-number requests can still produce native tracking. Detection is not disabled or falsified; unexpected historical issues47/48 remain evidence.
Tool-name allowlists are pre-approval, and safe-output targets are defaults rather than strict issue-number confinement. Evidence decisions and comment deduplication remain model-driven; writes are not transactional. Actual coding-agent assignment, positive real external-author handoff and forced live outages remain unproven. Shared runner capacity cannot be guaranteed unchanged despite preserving legacy workflow code and dependencies.
All SDK Contribution checklist