How it works

Inside your account. Rules decide. Proof verifies offline.

Wardhand replaces manual screenshots with a repeatable loop inside your own AWS account. Your data stays in your account. The drafting model runs in your own cloud by default; using a hosted model is your explicit choice.

Deterministic loop Every 15 min
5 STEPS IN YOUR VPC 1 Read metadata 2 Rules decide 3 Fix or PR 4 Live re-read 5 Offline seal

The five-step loop inside your AWS account

Fixed rules end to end
How it is known
Wardhand separates observation, judgment, drafting, and approval. No AI model ever passes a check or changes a cloud resource.

1. Read configuration metadata—never customer data

Read-only boundary

Every 15 minutes, Wardhand reads configuration metadata across your connected AWS accounts and GitHub repositories. It cannot read S3 objects or secrets, and stores hashed records in your own account.

15-minute sessions No S3 object reads Raw evidence kept

Kept as returned by AWS

CC6.6CC7.2

2. Fixed rules decide every check—guided by your policy

53 checks

All 53 checks across AWS and GitHub run by fixed rules mapped to SOC 2 criteria and your signed policy. If a read fails, Wardhand reports "could not test"—never a fake pass.

No model calls for checks Honest not-tested state Policy-driven thresholds

Decided by written rule

CC6.1CC9.1

3. Check impact first, then fix or open a pull request

Human signed

Before changing anything, Wardhand shows attached servers, networks, or policies. Approve a reversible AWS fix, or open a minimal Terraform pull request checked by fixed rules—never auto-merged.

Before & after diff Instant undo Peer-reviewed PRs

Signed human grant

CC8.1Undo kept

4. Live re-read seals the receipt—or closes outside fixes

Self-closing

After an approved fix, Wardhand re-reads AWS live and seals before-and-after proof into a signed receipt. If your team fixes or deletes the resource outside Wardhand, the next scan closes the finding.

Before & after proof Closes outside fixes Tamper-evident chain

Read back from system

ReceiptHash chain

5. Export one signed bundle your auditor verifies offline

Offline proof

Export checks, policies, access reviews, exceptions, and fix receipts in one signed bundle. Your auditor verifies it offline with your public key—no access to your AWS account required.

wardhand verify-workpaper Out-of-band public key

Tamper-evident chain

SignedOffline

One finding, every fact, one deliberate decision.

When a check or policy fails, Wardhand shows what is affected first. Choose a reversible AWS fix or a small Terraform pull request, with raw evidence kept underneath.

Decision workbench Human signed · Reversible
Production database backup retention Policy §6.2 gap
Your policy §6.235 days
Live database7 days
Impact check Attached servers, networks & policies Checked first
Path A · One-click reversible AWS fix Undo kept

Shows exact before and after settings. One signed click applies the fix, re-reads AWS, and seals a receipt.

Path B · Minimal Terraform pull request Never auto-merged

Opens a small pull request in GitHub for peer review. Fixed rules check every diff, and Wardhand never merges it.

Self-closing: fix it directly, merge the PR, or remove the resource. The next verified scan closes the finding.

Evidence your auditor can verify without trusting us.

The offline verifier checks the bundle against the public key and installation ID you provide. Change even one character, and verification fails immediately.

Offline verification · clean bundle
$ wardhand verify-workpaper \
    --bundle wardhand-auditor-export-2026-Q4.tar.gz \
    --public-key wardhand-public-key.json \
    --installation northwind-prod

{"ok":true,"records":249,"chain":"verified"}
Tamper check · one character modified Refused
$ wardhand verify-workpaper \
    --bundle wardhand-auditor-export-edited.tar.gz \
    --public-key wardhand-public-key.json \
    --installation northwind-prod

ERROR invalid_bundle_or_trusted_key
  signed export record digest mismatch

Every fact carries a mark saying how it is known.

The right-hand margin shows whether a line came from AWS, a live re-read, a human signature, a fixed rule, or a gap.

In the source's own log
Read back from the system
Signed by a person
Decided by a written rule
Seen in a kept observation
In Wardhand's record only
Done by Wardhand
Not proven
Does not hold

See the loop on your own AWS Organization.

Start read-only in 60 minutes. Zero writes, and your data stays in your account.