AI Agent Risk Report for Clients: An OpenClaw Template for Agencies
Published 2026-10-30 · Updated 2026-10-30 · By The Is My Agent Safe team
Short answer
A client risk report for an OpenClaw install should say what you checked and when, which version runs, what OpenClaw's own audit found by severity, what you did about each finding, which findings the client accepted and why, and what the report can't show. Build it from openclaw security audit --deep --json and openclaw --version, use OpenClaw's own severity levels instead of a made-up score, and date every line.
- Scope and date first: which Gateway, which host, who ran the checks, and when.
- Findings by severity, using OpenClaw's own checkId and severity, with an action, an owner and a date for each.
- Accepted findings with the client's reason, kept visible in the audit JSON.
- Version against OpenClaw's published security advisories.
- A plain limits section: a dated result, not a promise that nothing can go wrong.
Paste your audit: run openclaw security audit --deep and paste the output into our free check for a red/amber/green result with the most important finding first (Limits of the check).
Who this is for
Agencies and consultants who run OpenClaw for clients and need to hand over a written risk report: after a setup, at a monthly review, or when a client's board, insurer or customer asks. This page is about the report itself, the document the client reads. For the records behind it, see How to show clients your OpenClaw security work. For the one-off handoff audit, see How to audit a client's OpenClaw.
Generic AI risk templates ask about data, people and harms. They're useful for the business case. A report on a running OpenClaw install also needs facts you can only get from the install, and OpenClaw's own commands give you those.
Step 1: Collect the facts (read-only)
Run these on the client's Gateway host, as the user that runs the Gateway. None of them changes config.
openclaw --version openclaw security audit --deep --json > "audit-$(date +%Y%m%d).json" openclaw secrets audit --check; echo "exit code: $?" openclaw config get gateway.bind
- The audit JSON gives you every finding with a structured
checkIdand a severity. OpenClaw's CLI docs showjq '.summary'for the counts. --deepadds best-effort live Gateway checks that the plain audit doesn't run.openclaw secrets audit --checkreturns exit code 1 on findings and 2 for unresolved references, per the secrets CLI docs. Record the exit code, never the secrets.- Don't run
openclaw security audit --fixwhile you're collecting. Fixes come after the client has seen the findings.
Step 2: Write the scope line
One short paragraph at the top. A client reading only this line should know exactly what the report covers.
- Which install: the Gateway host (a label, not a hostname you wouldn't want shared), the OS user and the agents on it.
- When: the date and time the audit ran, with the time zone.
- Who ran it, and from where.
- What it doesn't cover. For example: "We didn't review the code of the 4 installed skills" or "We didn't test from outside the client's network."
Step 3: Summarize in plain words
Three to five sentences, written last. Say what you found and what happens next. Use the audit's own counts, such as "0 critical, 2 warn, 3 info".
Avoid words that promise an outcome. "No critical findings in OpenClaw's audit on 30 Oct" is a fact the client can check. "The agent is locked down" is a claim nobody can back.
Step 4: The findings table
One row per active finding, critical first. Keep OpenClaw's checkId so anyone can look it up in the audit checks catalog.
| checkId | Severity | What it means for this client | Action | Owner | Due |
|---|---|---|---|---|---|
gateway.bind_no_auth (made-up example) | critical | The Gateway answers on the network with no auth | Set bind to loopback, add token auth | Agency | 31 Oct |
tools.exec.security_full_configured (made-up example) | warn | The agent can run any host command | Set exec to deny or ask | Agency, with client sign-off | 7 Nov |
The rows above are made-up examples, not a real client. Write the middle column for this client: which channel, which data, which tool. That's the part a generic template can't do for you.
Use OpenClaw's severity, not a score. Two agencies would score the same install differently, and the client can't check a score. They can check a severity against the audit file.
Step 5: Accepted findings, kept visible
Some findings are deliberate choices, such as a plugin the client relies on. Record them in config with security.audit.suppressions and a reason. OpenClaw's CLI docs say suppressed findings leave the active summary but stay in the JSON under suppressedFindings, and an info finding, security.audit.suppressions.active, shows the audit was filtered.
In the report, list each one with the client's reason, who agreed it and when. Don't drop them from the report because they're no longer in the summary.
Step 6: Version and advisories
Write the version from openclaw --version and the date you checked OpenClaw's security advisories. Then either "none apply to this version" or the advisory IDs that do, with the version that fixes each.
Compare against the current stable release on OpenClaw's releases page on the day you write the report, not a number you remember. Versions before 2026.8.1 are missing fixes for advisories rated high.
Step 7: How the agent can be reached
Clients often don't know this part, and it's where much of the risk sits. Use the inventory from OpenClaw's exposure runbook:
- Gateway bind mode and port (18789 by default), and the auth mode.
- Every enabled channel, and whether it accepts DMs, groups or webhooks.
- Which agents non-local senders can reach, and the tool profile and sandbox mode of each.
- The external credentials those agents hold.
Add one line from OpenClaw's trust model: everyone who can message a tool-enabled agent shares that agent's delegated tool authority. If the client's staff and the public can both message the same agent, say so in the report.
Step 8: Limits and next date
End with a short limits section, in plain words:
- The report shows what OpenClaw's audit and your checks found on the date shown. It isn't a promise that the install can't be attacked.
--deeplive Gateway checks are best-effort, per OpenClaw's docs.- New releases and advisories keep coming, so the version section goes out of date.
- List anything you didn't check, such as skill code, MCP servers or backups.
Then give the date of the next review, and the events that bring it forward: an OpenClaw update, a new channel, a new tool or a new skill.
A one-page outline you can copy
Client: [client name] Install: [label] Audit run: [date, time, zone] Run by: [agency] Scope: [what is and isn't covered] Summary: [3 to 5 plain sentences] Findings: [table: checkId | severity | meaning | action | owner | due] Accepted findings: [checkId | client's reason | agreed by | date] Version: [openclaw --version] | advisories checked [date]: [none apply / IDs] Reachability: [bind, auth, channels, tools, credentials] Secrets audit: [exit code, date] Limits: [what this report can't show] Next review: [date, and the changes that bring it forward]
What our free check covers
We built a free paste check you can use on one client install before you write the report. Paste a client's openclaw security audit --deep output and openclaw --version. It puts the most important finding first, in plain English, with how to fix that one when we have steps for it. It also looks for risky settings, such as API keys handed to skills, and checks the version against published OpenClaw security advisories. Nothing is installed into the agent you're checking.
We check the text you paste. Your browser hides many common key and token formats before sending, but not all of them, so remove any other secrets first. Green means no issues found in the checks we ran. See Limits of the check.
The check is free, and a full report is $29. One-off $29, not a subscription. Refunded automatically if the check can't finish, or in full if you ask within 14 days. See Limits of the check. The full report explains every finding and lists the fixes in order. We don't review the code inside skills or MCP servers ourselves, beyond what OpenClaw's own audit reports.
Is My Agent Safe is independent and isn't affiliated with or endorsed by OpenClaw or its maintainers.
For agencies with several clients
We're holding 20 founding seats for agencies. The seat gives you a bulk check of up to 25 client installs (paste each install's OpenClaw audit output), re-checks of your clients against new OpenClaw advisories without pasting again, for 90 days from each install's last paste, and a PDF report for each client with your agency's name. 14-day refund on your first payment. It's for businesses located in the United States or the United Kingdom, bought by card. $99/month, price locked for your first 12 paid months. After that, it moves to the then-current Agency price, with at least 30 days' email notice. Cancel any time. It renews automatically every month until you cancel, and you can cancel online from your dashboard. Offer conditions.
See the seat and a sample client report: Is My Agent Safe for agencies.
Where next
- OpenClaw: openclaw security (CLI), Security audit checks, Gateway exposure runbook
- How to show clients your OpenClaw security work
- Weekly OpenClaw security checks for clients
- openclaw security audit explained
Frequently asked questions
What should an AI agent risk report for a client include?
Scope and date, a short plain-English summary, a findings table (checkId, severity, what it means, action, owner, date), accepted findings with reasons, the version and the advisories you checked, how the agent can be reached, and a limits section. For OpenClaw, each of these comes from OpenClaw's own commands.
Should I give the client a risk score?
Use OpenClaw's own severity levels (critical, warn, info) and counts from the audit summary. A home-made score out of 100 is hard to defend, and two agencies would score the same install differently. If the client wants one headline, use the count of open critical findings.
How do I show a finding the client chose to accept?
OpenClaw lets you accept a finding in config with security.audit.suppressions and a reason. Its CLI docs say suppressed findings leave the active summary but stay in the JSON under suppressedFindings, so the report can list them honestly.
How often should the report be updated?
Agree a date with the client, and re-run it after any OpenClaw update, new channel, new tool or new skill. A report is only as current as the day the audit ran.
Can Is My Agent Safe write the report for me?
Partly. We built a free paste check: paste a client's openclaw security audit --deep output and openclaw --version, and it puts the most important finding first, in plain English, with how to fix that one when we have steps for it. The founding agency seat adds a PDF report for each client with your agency's name. We don't review the code inside skills or MCP servers ourselves, beyond what OpenClaw's own audit reports.
Sources
- OpenClaw docs: openclaw security (CLI: JSON output, suppressions)
- OpenClaw docs: security audit checks
- OpenClaw docs: Gateway exposure runbook
- OpenClaw docs: security trust model
- OpenClaw docs: openclaw secrets (CLI)
- OpenClaw docs: openclaw backup (CLI)
- OpenClaw GitHub security advisories
- OpenClaw releases (GitHub)
Get the plain-English agent security brief
Agents & AI Security, explained: a short, plain-English email when we have something useful. Free. Unsubscribe any time.
We'll only use your email to send the newsletter. The newsletter is sent by beehiiv, which records opens and link clicks. Privacy policy.