Skip to content

Add /fix-issue workflow for automated issue fixing#1313

Open
vikrantpuppala wants to merge 2 commits intodatabricks:mainfrom
vikrantpuppala:claude-fix-issue-workflow
Open

Add /fix-issue workflow for automated issue fixing#1313
vikrantpuppala wants to merge 2 commits intodatabricks:mainfrom
vikrantpuppala:claude-fix-issue-workflow

Conversation

@vikrantpuppala
Copy link
Copy Markdown
Collaborator

Summary

  • Adds a new GitHub Actions workflow (claude-fix-issue.yml) that lets maintainers comment /fix-issue on any issue to trigger Claude to automatically reproduce, fix, test, and create a PR
  • Updates fix-github-issue.md skill with security guidance for treating issue content as untrusted data and CI non-interactive mode support
  • Security model: maintainer-gated trigger, hardcoded prompt treats issue body as data (not instructions), WebFetch/WebSearch blocked, audit trail comment posted before code changes begin

Test plan

  • Open a test issue on the repo, have a maintainer comment /fix-issue, verify the workflow triggers
  • Verify the author_association gate blocks non-maintainer triggers
  • Verify Claude posts a summary comment before making code changes
  • Verify Claude creates a PR with the fix (not just a branch link like [BUG]IS_NULLABLE field is incorrect for OSS JDBC metadata getColumns operation #1299)
  • Test with an issue containing prompt injection attempts (e.g., "ignore previous instructions and...") to verify they're ignored

NO_CHANGELOG=true

This pull request was AI-assisted by Isaac.

Adds a new GitHub Actions workflow that lets maintainers comment /fix-issue
on any issue to trigger Claude to automatically reproduce, fix, test, and
create a PR. Security-hardened: maintainer-gated trigger, issue body treated
as data not instructions, WebFetch/WebSearch blocked, audit trail comment
posted before code changes.

Also updates fix-github-issue.md skill with security guidance for treating
issue content as untrusted data and CI non-interactive mode support.

Co-authored-by: Isaac
Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>
@shivam2680
Copy link
Copy Markdown
Collaborator

Apply recommendations from the internal "Agentic Security Threats and
Mitigations" document: add non-override language for security rules,
spotlighting guidance for untrusted content, explicit prohibition of
external data exfiltration tools, and expanded prompt injection patterns.

Co-authored-by: Isaac
Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>
@samikshya-db
Copy link
Copy Markdown
Collaborator

/review

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants