Add LLM issue auto-triage: type and high-severity labelers#584
Merged
Conversation
Uses strands-agents/devtools/issue-labeler (pinned by SHA) on new issues. Two parallel jobs: type (bug/enhancement/question, max 1) and severity (high-severity or nothing, conservative criteria). The model has no tools or GitHub access; labels are schema-constrained to the config allowlist.
agentcore-devx-automation
Bot
added
the
claude-security-reviewing
Contributor
|
Claude Security Review: no high-confidence findings. (run) |
agentcore-devx-automation
Bot
removed
the
claude-security-reviewing
Dry-run against real issues showed ~36% of bug reports flagged, largely because detailed reports of narrow-path crashes matched the crash criterion. Add a blast-radius test: the failure must hit a core path a typical production user would exercise; crashes confined to optional third-party integrations with specific feature combinations are ordinary bugs. Verified on 14 real issues: flags import-breaking regression (#496), silent data loss (#564), microVM reaping (#471); no longer flags narrow Strands-integration crashes (#556, #583).
agentcore-devx-automation
Bot
added
the
claude-security-reviewing
Contributor
|
Claude Security Review: no high-confidence findings. (run) |
agentcore-devx-automation
Bot
removed
the
claude-security-reviewing
tejaskash
requested changes
Jul 21, 2026
tejaskash
left a comment
tejaskash
left a comment
Contributor
There was a problem hiding this comment.
Looks good, but not sure why we're pinning the strands reviewer to a certain commit.
tejaskash
enabled auto-merge (squash)
July 21, 2026 18:39
tejaskash
approved these changes
Jul 21, 2026
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
2 participants
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
Adds automatic issue triage on
issues: openedusing strands-agents/devtools/issue-labeler (the same setup strands-agents runs on their SDK repos), pinned by commit SHA.Two parallel jobs:
.github/labelers/type.yml): applies exactly one ofbug/enhancement/question..github/labelers/severity.yml): applieshigh-severityor nothing. Criteria are deliberately conservative — production outage, data loss, security vulnerability, crash with no workaround, or regression in a released version, and only with concrete evidence. Feature requests, usage questions, and vague reports never qualify. Body truncation is raised to 4000 chars for this job so stack traces/repro steps are visible to the classifier.The
high-severitylabel has been created on the repo.Security model
The classifier LLM has no tools, no shell, and no GitHub access. It returns a structured object whose label field is an enum built from the config allowlist, so out-of-allowlist labels are rejected by the schema. Issue content is sanitized and truncated before reaching Bedrock. IAM is scoped to
bedrock:InvokeModel*via an inline session policy; the workflow needs onlyissues: write. Worst case from prompt injection in an issue body is a mislabel.Notes for reviewers
secrets.BEDROCK_SECURITY_REVIEW_ROLE_ARN(same OIDC role as the PR security review). If that role's trust policy pins thesubclaim to pull-request refs, it will rejectissues-triggered runs — needs a check before merge, or a dedicated minimal role.5e50593(strands devtools main as of 2026-07-17); bump deliberately.high-severitylands, via the existinggithub-slack-notifications.ymlwebhook.