Skip to content

Gate agent workflows on trusted authors - #2443

Draft
devi-labs wants to merge 1 commit into
mainfrom
fix/agent-trusted-author-approval
Draft

Gate agent workflows on trusted authors#2443
devi-labs wants to merge 1 commit into
mainfrom
fix/agent-trusted-author-approval

Conversation

@devi-labs

Copy link
Copy Markdown
Contributor

Summary

Prevent write-capable agent workflows from running on externally authored issues or PRs, even when a trusted actor triggers them. Add GitHub Environment approval gates for the higher-risk agent jobs, with naqvitalha as the only approval-bypass actor for now.

Approach

  • Require issue/PR authors to be OWNER, MEMBER, or COLLABORATOR before agent workflows can run.
  • Add approval jobs for agent-fix, agent-bot, and agent-android-bot using GitHub Environments.
  • Keep existing trigger-actor permission checks, Shopify AI Proxy security scan headers, and untrusted-content prompt language in place.

Validation

  • ruby -e 'require "yaml"; ARGV.each { |f| YAML.load_file(f) }; puts "all workflow yaml ok"' .github/workflows/*.yml
  • git diff --check -- .github/workflows/agent-triage.yml .github/workflows/agent-fix.yml .github/workflows/agent-bot.yml .github/workflows/agent-android-bot.yml
  • dev check reports no checks configured for this repo.

Deployment note

The approval environments must be configured in GitHub repository settings with required reviewers for the approval gates to pause runs:

  • agent-fix
  • agent-bot
  • agent-android-bot

Co-authored-by: gpt-5.5 noreply@shopify.com
Orchestrated-by: ae noreply@shopify.com

## Summary

Prevent write-capable agent workflows from running on externally authored issues or PRs, even when a trusted actor triggers them. Add environment approvals for higher-risk agent jobs while allowing naqvitalha to bypass that approval path for now.

## Approach

- Require issue and PR authors to be OWNER, MEMBER, or COLLABORATOR before agent workflows can run.
- Add approval jobs for agent-fix, agent-bot, and agent-android-bot using GitHub Environments.
- Keep existing trigger-actor permission checks and prompt-injection scan headers in place.

Co-authored-by: gpt-5.5 <noreply@shopify.com>
Orchestrated-by: ae <noreply@shopify.com>
Assisted-By: devx/1eb1773d-0aa4-4609-9e57-479a175eed30
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.

1 participant