More screenshots
Works with the AI Agents you already use
cloud attack path analyzer
Model your AWS account as an attack graph: real IAM evaluation, 30+ techniques, ranked internet-to-crown-jewel paths, a minimal fix plan, min-cut, blast radius, SARIF/HTML reports.
$29
About this skill
Scanners give you 400 findings. Attackers only need one path. This finds the paths.
Single misconfigurations rarely sink a company. Chains do: an unscoped GitHub OIDC trust leads to a deploy role, which can PassRole to an admin Lambda, which can read the customer database. This skill builds that chain explicitly. It imports read-only AWS CLI exports (or a hand-written inventory), evaluates every identity and resource policy with a real IAM engine (wildcards, explicit Deny, NotAction/NotResource, permission boundaries, trust policies, conditions), and adds edges for 30+ attacker techniques. Those include IMDSv1 credential theft, exposed admin and database ports, exploitable CVEs, open or confused-deputy trusts, nine IAM privilege-escalation methods, PassRole to Lambda/EC2, user takeover, SSM remote execution, public buckets and function URLs. Graph algorithms then do the rest: Dijkstra plus Yen's k-shortest paths rank every route from the internet to admin and your crown-jewel data, a greedy hitting-set planner gives the smallest ordered list of fixes that breaks every MEDIUM-or-worse path (re-routing after each fix the way an attacker would), Edmonds-Karp computes the minimum cut, and a blast-radius table shows what falls if any single identity or host is compromised.
How it works
- Import: import-aws turns read-only AWS CLI exports into one inventory file (IAM users, roles, groups, managed and inline policies, boundaries, trust policies, credential report, EC2, security groups, Lambda, bucket policies), or you write the inventory by hand.
- Evaluate: an offline IAM engine answers "can X do Y on Z?" with explicit-Deny precedence, NotAction/NotResource, boundaries and conditions.
- Graph: 30+ attacker techniques become weighted edges between the internet, identities, hosts, functions, data and account admin.
- Rank: Dijkstra plus Yen's k-shortest paths find the easiest routes to every crown jewel, scored by difficulty × data sensitivity.
- Fix: a greedy hitting-set planner picks fixes by paths-broken per effort and re-routes after each one. A max-flow min-cut gives the smallest set of changes that disconnects every crown jewel.
- Report: text, JSON, Mermaid, self-contained HTML, SARIF 2.1.0, plus diff mode for CI.
About 1,500 lines of standard-library Python (a CLI plus 9 library modules), with 62 unit tests.
What it does
- Offline IAM evaluator: wildcards, explicit Deny, NotAction/NotResource, boundaries, trust and resource policies, conditional grants
- 30+ edge types: network exposure, CVEs, IMDSv1, OIDC/cross-account/open trust, 9 privesc methods, PassRole, user takeover, SSM, public data
- Ranked paths (Dijkstra + Yen's k-shortest) scored by difficulty × data sensitivity
- Ordered fix plan that re-routes after every fix until no MEDIUM+ path is left, plus Edmonds-Karp min-cut and choke points
- Blast radius per identity and host
- Reports: text, JSON, Mermaid diagram, self-contained HTML, SARIF 2.1.0 (GitHub code scanning). Diff mode and --fail-on for CI
- import-aws from read-only CLI exports, plus explain-policy for any single action
Who it's for
Cloud security engineers, DevSecOps teams, consultants doing AWS reviews, and startups preparing for a pen test or SOC 2.
Limits
Model-based: only as complete as the inventory. SCPs, session policies, VPC endpoint policies and KMS key policies are not modelled. Condition values aren't evaluated; conditional grants are weighted as harder, not removed. Only the AWS importer is included. Other clouds need a hand-written inventory.
Model-based analysis of the inventory you provide. It does not test or touch live systems and is not a penetration test. Paths depend on how complete the inventory is. Validate findings before changing production.
What's included
- SKILL.md: workflow, commands, how to explain results
- scripts/: the tool (standard-library Python)
- references/: sources, inventory format, scoring and edge catalog
- examples/: sample inputs used in the demo (all made up)
- tests/: unit tests (Python 3.8 / 3.9 / 3.13)
Standard-library Python only. No API keys, no network calls, no installs.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
Model-based: only as complete as the inventory. SCPs, session policies, VPC endpoint policies and KMS key policies are not modelled. Condition values are not evaluated; conditional grants are weighted as harder, not removed. Only the AWS importer is included. Other clouds need a hand-written inventory. Model-based analysis of the inventory you provide; it does not test or touch live systems and is not a penetration test.
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 buying a skill.
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 today