[alanshaw]

A bucket policy can now grant these by name: object tagging (s3:GetObjectTagging, s3:PutObjectTagging, s3:DeleteObjectTagging), s3:GetObjectAttributes, and reads of the bucket's configuration (s3:GetBucketAcl, s3:GetBucketCORS, s3:GetBucketLocation, s3:GetBucketObjectLockConfiguration, s3:GetBucketTagging, s3:GetBucketVersioning). The configuration reads are bucket actions, so they take the bucket ARN; the rest take the objects ARN.

Hilt is changing to refuse any request with a subresource it doesn't classify, and to give each operation the gateway serves its own permission. Before that change, a principal reached these operations through s3:ListBucket, s3:GetObject and the plain object writes, so policies need to be able to grant them or principals lose access.

The configuration writes (s3:PutBucketVersioning, s3:PutBucketObjectLockConfiguration, s3:PutBucketTagging) stay out of the allowlist, as the encryption configuration does: each changes the bucket for every principal. Service keys can hold them.

Policies using s3:* pick up the new actions automatically. Policies that list actions need the new ones added.

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The focused permission additions and validation tests match the intended scope, with no blocking issues identified.

No open issues 🎉

What changed in this PR

Expands bucket-policy permissions to support Hilt’s operation-specific authorization while keeping bucket configuration writes excluded.

Changes:

• Allows object tagging, object attributes, and six bucket configuration reads.
• Classifies configuration reads as bucket-ARN actions.
• Adds tests for ARN validation and rejected configuration writes.

File	Description
commands/​s3/​bucket/​policy_test.go	Tests resource restrictions and excluded configuration writes.
commands/​s3/​bucket/​policy_json.go	Extends the action allowlist and bucket-action classification.

_{Review effort: Balanced}

[pyropy]