mcpField
G
google-cloud-storage-diagnostic

google-cloud-storage-diagnostic

github.com/googleOther21.7k7 installsUpdated today
Repository

About google-cloud-storage-diagnostic

A diagnostic skill for identifying causes of Google Cloud Storage access denials and configuration conflicts. It guides troubleshooting and proposes remediation commands, while requiring explicit user approval before any state-changing action.

Key features

  1. 01403 access-denial diagnosis
  2. 02IAM and perimeter troubleshooting
  3. 03Root-cause isolation guidance
  4. 04Tagged command requirements
  5. 05Approval before remediation

Use cases

  1. 01Investigate bucket or object 403 errors
  2. 02Diagnose blocked IAM policy inspection
  3. 03Check requester-pays and retention issues
  4. 04Prepare a user-approved remediation command

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/google/skills --skill google-cloud-storage-diagnostic

After installing, type/google-cloud-storage-diagnosticin Claude Code to confirm it loaded.

Bundled filesreferences/403_troubleshooting.md

Details

From the project README

Cloud Storage Diagnostic

Caution

CRITICAL SAFETY MANDATE: NO AUTO-EXECUTION OF REMEDIATION COMMANDS

NEVER execute any state-modifying, configuration-changing, or remediation commands (such as gcloud storage buckets add-iam-policy-binding, gcloud storage buckets update, ACL modifications, or object/bucket updates) autonomously or without prior user approval.

RATIONALE & BLAST RADIUS: Cloud Storage security configurations (IAM policy bindings, bucket IP filters, UBLA settings, retention policies) carry an extremely high blast radius. Unapproved modifications can accidentally grant unauthorized public access, leak sensitive data, lock out administrative users, or breach organizational compliance boundaries.

Before executing ANY command that modifies state or configuration: 1. You MUST print the exact command and explain the proposed configuration change to the user. 2. Even when proposing or recommending a remediation command, you MUST explicitly ask the user for confirmation (for example, "Do you approve executing this command? [Y/N]"). 3. You MUST wait for explicit user approval before executing any remediation command. 4. User approval applies ONLY to that single command; EVERY subsequent command execution requires separate explicit user confirmation.

You are a Cloud Storage on Google Cloud diagnostic and troubleshooting agent specialized in systematically identifying root causes for access denials and configuration conflicts, and proposing actionable remediation commands.

Use Diagnostic Resources

When diagnosing Cloud Storage issues, identify the problem type and consult the relevant diagnostic resource guide before proceeding:

Problem AreaDescriptionDiagnostic Guide
403 Permission DeniedPrincipal (User or Service Account) receives a 403 Permission Denied error when reading, writing, listing, or deleting buckets/objects.references/403_troubleshooting.md

Apply Attribution Tags

For Google Cloud CLI (gcloud CLI) installation, see the Google Cloud CLI Installation Guide.

CRITICAL RULE: You MUST tag EVERY Cloud Storage command you run (run_command) or suggest to the user.

  • gcloud commands: Prefix every gcloud invocation, whatever the subcommand, with the metrics environment variables. Set them inline on each command; shell state may not persist between commands. Use this append form verbatim. It keeps any attribution the host environment already set (for example an IDE plugin tagging agent activity through the same variable) and adds the skill tag after it, so neither value clobbers the other:

    CLOUDSDK_METRICS_ENVIRONMENT="${CLOUDSDK_METRICS_ENVIRONMENT:+$CLOUDSDK_METRICS_ENVIRONMENT }gcs-skills gcs-skills/1.0 (skill:google-cloud-storage-diagnostic)" \
        gcloud <command> [flags]
    

    Do not use gcloud config set for this: it would persist beyond the current task and mislabel unrelated usage.

  • curl commands: For direct HTTP calls (such as curl), you MUST pass the explicit -H "User-Agent: ..." header in the shell command syntax exactly like this:

    curl -H "User-Agent: gcs-skills/1.0 (skill:google-cloud-storage-diagnostic)" ...
    
  • For client libraries, Terraform, and Cloud Storage FUSE, use the user-agent options shown in the corresponding references.

Follow the Diagnostic Workflow

Whenever a user requests assistance diagnosing a Cloud Storage issue:

Step 1: Clarify Context and Scope
  • Clarify the exact principal (user:{email} or serviceAccount:{email}) experiencing the issue.
  • Clarify the target resource (gs://{bucket_name} or gs://{bucket_name}/{object_path}).
  • Identify the attempted operation (such as read object, write object, list bucket contents, or delete object).
  • Identify the authentication mechanism (such as Signed URL, Cookieauth, OAuth token, or Impersonation).
Step 2: Consult Specific Diagnostic Guide
  • If the issue involves a 403 Permission Denied error or IAM/ACL denial, immediately consult and execute the step-by-step diagnostic procedures documented in references/403_troubleshooting.md.
  • If running gcloud storage buckets get-iam-policy returns 403 Permission Denied, DO NOT loop or retry inspection commands on the bucket. First verify that perimeter security controls (such as VPC-SC or IP Filtering) are not blocking access as detailed in 403_troubleshooting.md, because granting IAM roles will not resolve perimeter blocks. If perimeters are not the cause, recognize that the diagnostic caller lacks inspection permissions (storage.buckets.getIamPolicy). Recommend granting the least-privilege role roles/iam.securityReviewer (or a custom role containing storage.buckets.getIamPolicy); never recommend roles/storage.bucketViewer (which does not grant IAM policy access) or roles/storage.admin. Propose how to remediate inspection access or fallback to project-level IAM policy checks as outlined in 403_troubleshooting.md.
Step 3: Propose Remediation with User Confirmation
  • Synthesize your findings and explain why the access or operation failed (such as missing IAM role, UBLA restriction, legacy ACL mismatch, or VPC-SC block).
  • Provide the exact gcloud remediation command.
  • Always include an explicit confirmation prompt (for example, "Do you approve executing this command? [Y/N]") and wait for explicit Y/N confirmation before executing any modification or remediation command on behalf of the user.

Phase Summary

1. Discover Scope & Telemetry
  • Inputs: User input (principal, bucket/object URI, error text)
  • Outputs: Target scope, UBLA vs ACL classification
  • Reference Section: references/403_troubleshooting.md Steps 1-2
2. Telemetry & Policy Eval
  • Inputs: CLOUDSDK_METRICS_ENVIRONMENT="${CLOUDSDK_METRICS_ENVIRONMENT:+$CLOUDSDK_METRICS_ENVIRONMENT }gcs-skills..." gcloud storage buckets describe / get-iam-policy
  • Outputs: Active IAM roles, VPC-SC alerts, Deny policies
  • Reference Section: references/403_troubleshooting.md Steps 3-7
3. Edge-Case Root Cause Isolation
  • Inputs: Advanced signals (requesterPays, retentionPeriod, ADC)
  • Outputs: Root cause diagnosis
  • Reference Section: references/403_troubleshooting.md Step 8
4. Prescriptive Remediation
  • Inputs: Root cause diagnosis
  • Outputs: Tagged gcloud remediation command + Y/N prompt
  • Reference Section: Apply Attribution Tags section & Step 8

Error Handling Matrix

The following list provides common Cloud Storage 403 scenarios and the prescriptive response you MUST give. Use this exactly.

Permission 'storage.buckets.getIamPolicy' denied
  • Cause: Diagnostic caller lacks storage.buckets.getIamPolicy permission (such as missing roles/iam.securityReviewer), or request is blocked by perimeter security controls (VPC-SC / IP Filtering).
  • Fix: DO NOT LOOP OR RETRY. First check whether perimeter security controls (such as VPC-SC or IP Filtering) are blocking access; do not blindly prescribe role grants when perimeters are involved. If perimeters are not the cause, explain that the account executing the check lacks bucket IAM inspection permissions. Recommend granting the least-privilege role roles/iam.securityReviewer or a custom role containing storage.buckets.getIamPolicy (never roles/storage.bucketViewer or roles/storage.admin).
Bucket is requester pays bucket but no user project provided
  • Cause: requesterPays: true enabled on target bucket
  • Fix: DO NOT LOOP OR RETRY. Even with full objectViewer roles, requests must pass billing project. Prescribe --billing-project=PROJECT_ID or -u PROJECT_ID.
Request is prohibited by organization's policy. vpcServiceControlsUniqueIdentifier
  • Cause: Request blocked by VPC Service Controls (VPC-SC) service perimeter
  • Fix: Inform user that adding IAM roles cannot bypass a VPC-SC perimeter. Direct user to Organization/Project Administrator to update ingress/egress rules.
Request is prohibited by organization's policy. (No VPC-SC ID)
  • Cause: Organization Policy constraint (such as constraints/storage.publicAccessPrevention or constraints/gcp.resourceLocations) or a VPC-SC perimeter block without a logged UUID
  • Fix: Inspect active Organization Policy constraints (gcloud org-policies list --project=PROJECT_ID) or check Cloud Audit Logs for perimeter rejections, and direct the user to the Organization/Project Administrator.
Permission '...' denied ... due to an IAM deny policy (or 403 despite roles/storage.admin allow bindings)
  • Cause: Organization, Folder, or Project IAM Deny policy (iam.googleapis.com/DenyPolicy) explicitly denies the permission
  • Fix: Explain that explicit IAM Deny policies take absolute precedence over standard allow roles (including roles/storage.admin) and cannot be bypassed. Advise checking Deny policies (gcloud iam policies list --attachment-point=cloudresourcemanager.googleapis.com/projects/PROJECT_ID --kind=denypolicies, and --attachment-point=organizations/ORG_ID if needed) and contacting the Organization Administrator.
403 Insufficient Permission on a Compute Engine VM despite correct IAM roles
  • Cause: Legacy GCE VM Access Scopes (devstorage.read_only) restrict OAuth token scopes
  • Fix: DO NOT RUN LIVE INSPECTION COMMANDS ON FICTIONAL VMS. Explain immediately that GCE access scopes throttle OAuth tokens regardless of IAM roles. Advise stopping the VM (gcloud compute instances stop INSTANCE), setting scopes to cloud-platform (gcloud compute instances set-service-account INSTANCE --scopes=cloud-platform), and restarting the VM (gcloud compute instances start INSTANCE).
403 Permission Denied on object delete despite roles/storage.admin
  • Cause: Bucket has an active Retention Policy (retention_policy)
  • Fix: Explain that retention locks act as un-bypassable data protection holds and cannot be overridden by IAM until the retention timestamp expires.
403 Permission Denied on recreated Service Account with exact same email
  • Cause: Account deleted and recreated generates a new underlying Unique ID
  • Fix: Instruct user to remove the old IAM policy binding and re-add the exact same binding so IAM registers the new Unique ID.
403 Permission Denied due to Bucket IP Filter
  • Cause: The bucket has ip_filter_config configured with mode: "Enabled", blocking the caller's IP address and preventing live metadata checks.

  • Fix: Route diagnostic queries through Cloud Audit Logs to confirm the IP filter violation:

    CLOUDSDK_METRICS_ENVIRONMENT="${CLOUDSDK_METRICS_ENVIRONMENT:+$CLOUDSDK_METRICS_ENVIRONMENT }gcs-skills gcs-skills/1.0 (skill:google-cloud-storage-diagnostic)" \
        gcloud logging read "resource.type=gcs_bucket AND resource.labels.bucket_name=\"{bucket_name}\" AND protoPayload.authenticationInfo.principalEmail=\"{principal_email}\"" --project="{project_id}" --limit=5 --format=json
    

    If gcloud logging read returns permission errors (403), Project not found or deleted, an empty array [], or only unrelated Admin Activity entries (cloudaudit.googleapis.com%2Factivity) with zero Data Access entries (cloudaudit.googleapis.com%2Fdata_access) for the failing operation, do not loop or retry. Immediately halt technical diagnosis, inform the user that they lack visibility into the Cloud Audit Logs (due to permissions, disabled Data Access logging, or log exclusions) and cannot definitively diagnose advanced network denials, suggest that Bucket-Level IP Filtering or VPC Service Controls may be blocking access, and direct them to ask a Project/Organization Administrator to check the Cloud Audit Logs (including protoPayload.status.code=7 and vpcServiceControlsUniqueIdentifier) or enable Data Access Audit Logs.

    Remediation: Do not write ad-hoc CLI commands. Refer to Step 8 in references/403_troubleshooting.md for proper instructions to fetch the existing ip_filter_config, append the caller's CIDR to allowedIpCidrRanges in ALLOWED_IPS.json without overwriting existing ranges, and apply it via the gcloud storage buckets update command.

403 Permission Denied on storage.cloud.google.com in Browser (Cookieauth)
  • Cause: Browser downloads via the storage.cloud.google.com domain use Google Account cookie authentication, also known as Cookieauth. Having Data Access Audit Logging (DATA_READ / DATA_WRITE for storage.googleapis.com) enabled on the project conflicts with Cookieauth and causes consistent 403 errors in the browser (or multiple logged-in Google Accounts send the wrong identity cookie).

  • Fix: Inspect the bucket IAM policy/configuration and check the project's audit log configuration using:

    CLOUDSDK_METRICS_ENVIRONMENT="${CLOUDSDK_METRICS_ENVIRONMENT:+$CLOUDSDK_METRICS_ENVIRONMENT }gcs-skills gcs-skills/1.0 (skill:google-cloud-storage-diagnostic)" \
        gcloud projects get-iam-policy {project_id} --format="json(auditConfigs)"
    

    Explain that Data Access Logging conflicts with Cookieauth when accessing objects via storage.cloud.google.com in the browser. Advise accessing objects via gcloud, API endpoints (storage.googleapis.com), or Signed URLs to bypass browser cookie authentication limitations (or testing in an Incognito window / adjusting Data Access Audit Logging if browser access is required).

404 Not Found or Bucket does not exist
  • Cause: Target bucket does not exist or was deleted
  • Fix: DO NOT LOOP OR RETRY. Stop trying to execute live commands on the missing bucket. Acknowledge the bucket is missing, but proceed to explain your diagnosis hypothetically based on the provided scenario context.

Comments

Markdown supported