Works with the AI Agents you already use
CRA Risk Analyzer Pro
Enterprise-grade vulnerability analysis with vendor backport verification, MITRE ATT&CK mapping, and Monte Carlo risk pricing.
$5
CRA Risk Analyzer Pro
Example session with this skill installed
Analyse this Nessus report with 8,000 findings. Map the public attack chains for the Critical CVEs and give me a P90 loss estimate for the whole estate.
- Read your context and instructions
- Compiled the cra risk analyzer
I have processed the scan report and split it into 4 host-based parts for analysis. I mapped the 12 Critical CVEs against MITRE ATT&CK tactics and calculated the financial risk using a PERT distribution.
Scan Analysis & Risk Assessment
| Metric | Value | Details |
|---|---|---|
| Total Findings | 8,422 | Split into batch/parts/scan-part1.nessus to part4 |
| Critical CVEs | 12 | Primary drivers for ALE calculation |
| Confirmed Vulnerable | 8 | Validated against Red Hat & Ubuntu backport feeds |
| Attack Vector | Initial Access | Mapped to T1190 (Exploit Public-Facing Application) |
| P50 Expected Loss | $42,500 | Median annual loss estimate |
| P90 Expected Loss | $128,400 | High-end risk (10% probability of exceeding) |
| ALE (FAIR) | $51,200 | Annualised Loss Expectancy across estate |
Public Attack Chains (CVE-2023-4911)
- Tactic Mapping: Execution -> Persistence -> Privilege Escalation.
- Gaps: Attacker requires local shell access (T1059.004).
- Candidates: Public exploit available in CISA KEV; GitHub PoC observed.
- Vendor Status: Fixed in RHEL 8.8 (RHSA-2023:5455).
Next steps
- Review the generated update plan in
updates/2023-10-27-full-scan.md. - Run the generated Python script in
models/monte-carlo-20231027.pyfor a deeper sensitivity analysis. - Export these findings to HTML for the security steering committee.
Connects securely to your tools. The creator never sees your data.
About this skill
CRA Risk Analyzer Pro is the paid edition of CRA Risk Analyzer. It adds twelve Markdown skills for Claude Code, Codex and GitHub Copilot (agent mode) on top of the free CRA Risk Analyzer listing, and goes deeper in each of its three modes:
- One CVE — public exploit chains (CWE -> CAPEC -> MITRE ATT&CK, GitHub Advisory, OSV, Red Hat, CISA KEV) added automatically, and on request a loss estimate without any host assessment: PERT, annualised loss expectancy and P90 / P99.
- A CVE list for one host — attack trees built from the chains found on the host, full probability distributions and sensitivity, an update plan with vendor backport checks, and JSON / HTML / XML export with each CVE's exploitability, chain partners and the host's risk.
- A full scanner report — no 50-CVE cap: a report above 5,000 findings or 20 MB is split per host, the fetch time is shown before any lookup, one bulk survey covers groups of hosts, every host gets its own assessment, and chains that cross from one host to another are found.
What you get
- cra-pro-probability: the full distribution of how many hosts of a CVE may be compromised (P(X >= k)), the Poisson count of successful events per host now, at the due date and over a year, sensitivity of the yearly probability and expected loss, and attack trees with alternative (OR) and required (AND) steps.
- cra-pro-pricing: three-point loss estimates, the PERT mean, a FAIR-style annualised loss expectancy shown next to the free edition's expected loss, and loss percentiles (P50 / P90 / P99).
- cra-pro-codegen: generates a self-contained, seeded local model (Python, Go, PowerShell, bash or C#; standard library only, no network) for heavy computations such as Monte Carlo, only when you agree, and checks it against a worked example before using its results.
- cra-pro-chaining: public chain discovery: CWE -> CAPEC -> MITRE ATT&CK mapping, the tactics an attacker still needs, public candidates from NVD and CISA KEV, GitHub Advisory, OSV and Red Hat enrichment, with your own inventory chains ranked first and every row labelled observed, documented or inferred.
- cra-pro-export: JSON, single-file HTML (no scripts, no external requests, opens offline) and XML versions of a free-edition report.
- cra-pro-fixes: a per-host update plan from a scan. It keeps the findings at or above a severity threshold (High by default), checks each host against its distribution's vendor data and language packages against the upstream ranges, and lists per component one target version that closes all its CVEs, the CVEs it closes and their highest severity, the pairs already fixed (including backport-fixed) and the pairs that cannot become an update, with the reason. Advisory only: it installs nothing.
- cra-pro-batch: full scanner lists without the free edition's 50-CVE cap. It splits a report above 5,000 findings or 20 MB (CSV or Nessus XML) into per-host parts the free import reads, shows how many CVE cards must be fetched and how long that takes with and without an NVD API key (and asks above 200), runs one bulk survey where you answer once per group of hosts, writes one assessment per host, and finds chains that cross from one host to another it can reach.
- cra-pro-vendor-redhat, cra-pro-vendor-debian, cra-pro-vendor-ubuntu, cra-pro-vendor-suse, cra-pro-vendor-oracle: vendor fix data for a CVE card from each vendor's structured security data: per release or product and package, the fix state, fixed build, erratum (RHSA, USN, SUSE-SU, ELSA) and date. Vendors backport fixes without raising the upstream version, so these rows tell a fixed host from a vulnerable one where a version-only check cannot. Debian's structured feed carries no DSA / DLA advisory IDs, so Debian rows have no advisory.
How it works
- Assess with the free edition:
cra-setup,cra-cve,cra-scan,cra-assessandcra-reportproduce the cards, assessments and reports Pro builds on. - Go deeper:
cra-pro-probabilityandcra-pro-pricingappend their own sections to an assessment; for large or random computations they hand off tocra-pro-codegen. - Model locally:
cra-pro-codegenwrites a seeded model and its parameter file, runs it only with your consent, and passes results on only after its self-test passes. The same seed reproduces the same numbers. - Look outward:
cra-pro-chainingadds aPublicsubsection to a CVE card's chaining. - Plan updates:
cra-pro-fixesfilters a scan by severity, calls the vendor skill of each host's distribution, and writes the update plan per host. - Scale:
cra-pro-batchruns a whole scanner report host by host through the free import, survey and assessment, then hands the per-host files tocra-report. - Share:
cra-pro-exportcopies a finished report into JSON, HTML and XML.
Pro never changes a number the free edition computed. Every result is a Markdown, model or export file in your project workspace, with its inputs next to it, so you can recompute each number by hand. EXPOSED-RCE and OVERDUE flags stay visible everywhere.
Privacy & permissions
- Network: only
cra-pro-chainingand the vendor skills (called directly or bycra-pro-fixes), only when web access is available, and only to public sources.cra-pro-chaining: GitHub Advisory Database and Red Hat Security Data (the CVE ID), OSV (the GHSA advisory ID or the CVE ID), NVD (the vendor and product part of a CPE taken from the CVE card, never from a scan), CISA KEV (the public catalog is downloaded whole; nothing is sent), and the MITRE CWE, CAPEC and ATT&CK pages (the CWE, CAPEC or technique ID). Vendor skills: Red Hat Security Data and the Ubuntu CVE API (the CVE ID), SUSE's CSAF security files (the CVE ID and SUSE-SU advisory IDs), and the Debian security tracker and Oracle Linux OVAL feeds (bulk files downloaded whole; nothing is sent), fetched only after the severity filter ofcra-pro-fixesor for up to 5 CVEs you name, after a question about the size. If a CVE card is missing or stale, these skills — andcra-pro-batch, which makes no request of its own — first run the freecra-cvelookup, which sends the CVE ID to NVD and FIRST EPSS. Every outbound URL is shown in the conversation. - Never sent: host names, IP addresses, asset names, user names, file paths or scan rows.
- Environment variables: optional
GITHUB_TOKENandNVD_API_KEY, checked for presence only and used only as the auth header of GitHub Advisory and NVD requests. Their values are never shown, logged or stored. - Files written: only inside the workspace folder (default
cra/): sections appended to CVE cards and assessments, models and their parameter files undermodels/, exports next to the report they came from, update plans underupdates/, batch plans underbatch/(a large report is split into temporary part files there, deleted at the end of the run), and downloaded vendor files undercache/(reused for 7 days). Pro never changes a free-edition number. A re-run replaces only Pro's own parts: the## Advanced probabilityand## Advanced pricingsections, a card's### Publicsubsection and vendor blocks, and its own source rows andsourcesentries. An existing export is replaced only with your consent, and an export that fails its check may be renamed to.failed. The KEV download may use a temporary file that is deleted at the end of the run. - Commands run:
cra-pro-codegenwrites a standard-library model and runs it with a runtime you already have, only after you agree to generate it and again to run it. The model makes no network call, reads only its parameter file and writes no files; Go and .NET use their own build cache. On a .NET SDK older than 10, the model is a small project folder inside the workspace (undermodels/, named after the model and date), wheredotnet runwritesbin/andobj/.cra-pro-exportoffers a parse check of its files and runs it only with your consent.cra-pro-chainingfetches with curl, PowerShell or your agent's web-fetch tool; the vendor skills download with curl or PowerShell intocache/and read every response with a short inline Python 3 filter from their reference files that prints only validated structured fields (no script file ships);cra-pro-batchsplits a large report with a short inline Python 3 program that prints only counts, column names and the scan date;cra-pro-fixesruns no command on any host and prints no package-manager command. Nothing is installed. - Untrusted text: advisory, vendor, KEV, MITRE and report text is treated as data, never as instructions; vendor skills read only structured fields, never descriptions or notes; exports escape every value.
- Telemetry: none.
Does not perform active scanning or auto-patching. Local script execution requires a Python or PowerShell runtime. External vendor APIs require internet access.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 9 days ago
Requires