One wildcard in an IAM policy can turn a CI token into a full administrator, and one missing condition on a GitHub Actions role lets any workflow on github.com assume it. This skill finds those and seven other IAM mistakes before you deploy, and tells you how to fix each one.

Your agent runs a scanner on your policy files: standalone policy JSON, S3 bucket and KMS key policies, role trust policies, CloudFormation and SAM templates in YAML or JSON, CDK output and policies kept as JSON strings (Terraform .tf.json). It understands CloudFormation tags such as !Ref, !Sub and !If, IAM wildcards in actions (iam:Put*) and case-insensitive names. It never calls AWS. It returns BLOCK, REVIEW, or NOT FLAGGED, with the file, line, reason, and a safer approach.

What it checks

BLOCK (fix before shipping)

• I01: every action ("*") on every resource, or AdministratorAccess attached

• I02: Principal "*" with no condition that limits who can call

• I03: privilege escalation (iam:PassRole, iam:AttachRolePolicy, iam:CreatePolicyVersion and similar) on every role or user

• I04: a GitHub Actions, GitLab, EKS or Cognito trust with no sub or aud condition

REVIEW (check before release)

• I11: a whole-service wildcard such as s3:* or dynamodb:*

• I12: Allow with NotAction or NotResource

• I13: Principal "*" reading objects (a public bucket)

• I14: a whole AWS account trusted with no condition (no sts:ExternalId)

• I15: secretsmanager:GetSecretValue, ssm:GetParameter or kms:Decrypt on every resource

What's included

• The scanner (Python 3.9+, standard library only)

• A rules guide with the reason and safer approach for each rule

• Risky and safe example policies and CloudFormation templates, and the real report the scanner produces on the risky ones

Limits

It reads policy text as written, so it does not compute effective permissions: permission boundaries, SCPs, values built by CloudFormation functions and Terraform HCL (.tf) files are not evaluated. NOT FLAGGED is not a guarantee.

Contents

One ZIP file with 12 files: SKILL.md, the scanner in scripts/ (3 Python files), references/rules.md (every rule with its reason and safer approach), and examples/ (6 example files plus report.md, the real scanner output on the risky example).

How to use: This is an agent skill: a folder with a SKILL.md file and Python scripts that an AI coding agent such as Claude Code, Cursor, or Codex CLI can run. Unzip it into your agent's skills folder (for Claude Code: ~/.claude/skills/), then ask your agent to check your project. The scripts also run on their own with Python 3.9 or newer, with nothing to install. They read your files only: no network access, and nothing is sent anywhere.