The policy workbench: how it reads an AWS policy
Facts read 16 Nov 2026. Page reviewed 16 Nov 2026.
The workbench reads an AWS policy in your browser against a snapshot of AWS's service reference fetched on 16 Nov 2026: 457 services and 22,556 actions. It expands wildcards, labels access levels and runs six checks. It does not evaluate a policy against an account, and it uploads nothing.
Open the workbench ยท Download the reference snapshot
The reference data is the service reference that AWS publishes for automation, fetched on 16 Nov 2026. It lists 457 services and 22,556 actions, and for each action its access level, the resource types it can be scoped to and its condition keys. The workbench matches action names without regard to case, as AWS does, and keeps the fetch date with every count it prints. When AWS adds an action the count changes, so a count is only meaningful with its date.
Each action carries one of AWS's five access levels: List, Read, Write, Permissions management or Tagging. For an Action value the workbench expands the pattern against the reference: a star matches any run of characters, a question mark matches one, and a bare star matches every action. The count for a statement is the number of distinct actions after expansion, shown beside the number per access level. NotAction is read as every action except those listed.
Six checks run on the text, and each links to the AWS page that explains it. They report write or permissions management actions on Resource star; iam:PassRole on every resource; an Allow statement that uses NotAction; action names that match nothing in the reference; a trust policy that trusts another account, a star or a service with no guard condition; and a service control policy or resource control policy over its character limit, or a service control policy that carries a Principal element. The checks read the policy text only. They do not say whether a request would be allowed in your account.
The six checks and the AWS page behind each
| Check | Flags when | AWS documentation |
|---|---|---|
| Write or permissions management on every resource | An Allow statement grants Write or Permissions management actions with "Resource": "*" | best practices |
| iam:PassRole on every resource | An Allow statement grants iam:PassRole (directly or by wildcard) on "*" | passing a role |
| NotAction with Allow | An Allow statement uses NotAction | NotAction |
| Trust with no condition | A trust policy trusts another account or "*" with no sts:ExternalId, aws:PrincipalOrgID, aws:SourceAccount or similar condition | third-party access, confused deputy |
| SCP size | A service control policy is over 10,240 characters without white space (RCP: 5,120) | Organizations quotas |
| Elements an SCP does not accept | A service control policy statement has Principal or NotPrincipal | SCP syntax |
Every action name is also checked against the reference; a name that matches nothing is flagged, because AWS ignores it and the policy then grants less than its author meant. The checks are read from AWS's documentation on 16 Nov 2026; they are not IAM Access Analyzer itself, which runs inside an AWS account.
Twenty known cases
The workbench is checked against twenty cases kept in the site's tests folder; the build fails if one gives a different answer.
Cite this line: 'IAM Policy Generator, the policy workbench, edition 16 Nov 2026, https://iampolicygenerator.com/methodology/'
Related pages
Sources
- AWS Service Authorization Reference, programmatic access (fetched 16 Nov 2026), read 16 Nov 2026
- IAM Access Analyzer policy check reference, read 16 Nov 2026
- AWS Organizations quotas, read 16 Nov 2026
Last reviewed 16 Nov 2026.