aws-security

Plan least-privilege AWS access for an AI coding agent from a short spec, producing the trust policy, task-scoped permissions, a permission boundary, an SCP backstop, the assume-role command and a kill switch, or review an existing agent role against the same rules. Use when asked to "give my coding agent access to our AWS account", when an agent will deploy or run anything there, or when auditing an agent role. Not for human IAM users (iam-least-privilege-review) or organization-wide SCP design (scp-guardrails).

Audit one AWS account for security gaps by collecting inventory with read-only aws CLI commands and evaluating 17 checks offline (root MFA and keys, CloudTrail, GuardDuty, S3 public access, open security groups, old keys, admin policies, encryption), with severity, evidence and a fix per finding. Use when asked "is this account secure?", to baseline an account, before a handover or after an incident. Not for compliance evidence rows (aws-identity-and-logging-evidence), organization design (landing-zone-blast-radius) or deep IAM analysis (iam-least-privilege-review).

Audit what an AI coding agent did in an AWS account and which permissions it never needed, from a saved CloudTrail export filtered to the agent's role session: actions by service, writes versus reads, calls outside a declared allow list, logging tampering, destructive and IAM calls, resources touched, errors and the time window. Use when asked "what did the agent do in our AWS account?", after an agent session or before widening its role. Not for reviewing policy text (iam-least-privilege-review) or for an agent's local tool calls (agent-session-log-review in agent-security).

Write a filled-in AWS incident response runbook for a leaked access key, a compromised EC2 instance, a public S3 bucket, suspicious IAM activity, ransomware against S3 or EBS, or crypto-mining, with read-only inventory first, containment commands that each need confirmation, evidence preservation, recovery and templates; a triage mode picks the scenario from GuardDuty findings. Use when asked "we leaked an AWS key, what now?", when a GuardDuty finding needs a response plan or to prepare runbooks early. Not for a posture backlog (security-hub-triage) or routine audits (aws-account-audit).

Generate AWS Budgets, a Cost Anomaly Detection monitor and a sandbox SCP that denies expensive services, then check an exported Cost Explorer report for spikes. Use when asked to "stop runaway AWS spend", when setting cost alerts, protecting sandbox or agent accounts from runaway spend, or explaining a cost spike. Not for rightsizing or savings recommendations.

Review AWS IAM policies offline for over-broad permissions and privilege-escalation paths (full admin, service and action wildcards, unscoped iam:PassRole, NotAction in Allow statements, known escalation combinations), ranked by severity with a tightened policy per document. Use when asked "is this least privilege?", before approving a permissions change, or after an audit flags an admin policy. Not for SCPs (scp-guardrails) or resource policies such as bucket or key policies.

Design an AWS Organizations landing zone with one account per workload and environment, and show what a compromise of each account can reach. A bundled script turns a workload list into the OU tree, account names, foundation accounts and where SCP guardrails attach. Use when asked "how should we split our AWS accounts?", when planning a new organization, splitting a shared account or adding a workload. Not for auditing an existing account (aws-account-audit) or writing SCP JSON (scp-guardrails).

Generate a guardrail pack for an AWS sandbox OU used by engineers and AI agents, with linted SCPs (region allowlist, IMDSv2, no root, protected logging), spend budgets, a baseline checklist and a tag-based auto-expiry design. Use when asked to "set up a safe sandbox account", or when creating or tightening a sandbox or training OU. Not for production OUs (scp-guardrails) and it never runs the cleanup.

Build AWS Organizations service control policies from a short spec (region allowlist, break-glass roles, protected logging and detection, no root user, IMDSv2, no public S3 ACLs) packed under the 5120-character limit, and lint any SCP for Allow statements, region denies that break global services, NotAction misuse and size. Use when asked "why did our region SCP break IAM?", or when writing, reviewing or attaching an SCP. Not for IAM identity policies (iam-least-privilege-review) or the OU layout (landing-zone-blast-radius).

Turn a Security Hub or GuardDuty findings backlog into an owner-assigned next-actions list, offline from exports. A bundled script drops resolved and suppressed findings, groups the rest by severity, control and resource, assigns owners from rules and orders actions by severity and reach. Use when asked "what do we fix first in Security Hub?", for a weekly security review, or to route findings to teams. Not for running new checks (aws-account-audit) or responding to one active incident (aws-incident-response-runbook).