The AI Policy workflow (ai-policy.yml, check-ai-trailers job) fails on fork PRs when trying to apply the "AI assisted" label (ex: https://github.com/nextcloud/server#60070).
It 403s on both the label-create and label-assign calls because GitHub forces a read-only token (and withholds secrets) for pull_request-triggered runs from forks. The job's issues: write permissions and COMMAND_BOT_PAT fallback don't help.
The labeling step also has no || true, so it hard-fails the whole job, including the real policy check; even when there's no actual violation.
Suggestion: split workflow into two. Keep detection/policy-failing logic on pull_request (read-only), move the label-writing step to a separate pull_request_target-triggered job/workflow.
The AI Policy workflow (ai-policy.yml, check-ai-trailers job) fails on fork PRs when trying to apply the "AI assisted" label (ex: https://github.com/nextcloud/server#60070).
It 403s on both the label-create and label-assign calls because GitHub forces a read-only token (and withholds secrets) for pull_request-triggered runs from forks. The job's issues: write permissions and COMMAND_BOT_PAT fallback don't help.
The labeling step also has no
|| true, so it hard-fails the whole job, including the real policy check; even when there's no actual violation.Suggestion: split workflow into two. Keep detection/policy-failing logic on pull_request (read-only), move the label-writing step to a separate pull_request_target-triggered job/workflow.