Sightline: find the path, not the pile
Scanners hand you a pile of findings. Here is how to rank them by the path from the internet to your data, by hand or with Sightline.
Rank security findings by the route an attacker would take to your sensitive data, then fix the cheapest step on that route. A medium finding that sits on a short path to your payments database matters more than a critical one on a service that touches nothing. You can do this on a whiteboard in an afternoon. Sightline, which we are launching for SaaS and fintech teams without a security department, does it on every scan.
Why the pile does not help
Run a handful of open-source scanners over one repository and one domain and you can easily get hundreds of results. Each tool prints its own severity. Several tools report the same problem. Some results are safe by design, such as a public key that is meant to be public.
Severity scores do not settle it either. FIRST, which maintains CVSS, says in its user guide that base scores "are designed to measure the severity of a vulnerability", and that the base score should be supplemented with analysis of your environment. Severity describes the bug. It does not know where the bug sits in your system.
Two public feeds help with the "is anyone exploiting this" question:
- CISA's Known Exploited Vulnerabilities catalog. CISA says organisations "should use the KEV catalog as an input to their vulnerability management prioritization framework".
- FIRST's EPSS. EPSS estimates the probability that a published CVE will be exploited in the wild in the next 30 days, updated daily and free to use.
Both are about the vulnerability in general. Neither tells you whether it can reach your data. That part you have to work out.
What an attack path is
An attack path is a chain of steps from somewhere an attacker can start to something you cannot afford to lose. Each step is something the attacker can reach, use or become.
- Entry points. Anything reachable from the internet: your app, your API, old staging sites, a storage bucket set to public.
- Crown jewels. Where the sensitive data lives: card tokens, identity documents, personal data, credentials.
- Links. What lets an attacker move: an exposed route, a leaked key that still works, a service with permission to read a database.
Ranking by path means asking two questions of every finding. Is it on a path to a crown jewel? How many steps from the internet is it?
Map your paths by hand this week
You do not need a tool to start. Timings are a suggested starting point.
- List your crown jewels (30 minutes). Every database, bucket or third-party system holding sensitive data. Be specific: "payments-db, card tokens and customer emails".
- List your entry points (30 minutes). Every domain and subdomain, every public API, every bucket or file share with public access. Search your DNS records for old ones.
- Draw the links (1 hour). For each entry point, what does it talk to? What credentials does that service hold? What can those credentials read?
- Place your open findings on the drawing (1 hour). A leaked key goes on the service that holds it. An outdated package goes on the service that uses it.
- Count the steps. For each crown jewel, find the shortest route from an entry point. Write the number of steps next to it.
- Pick the cheapest link to cut. Often one change breaks the whole route: rotate a key, remove a route, tighten one permission.
| Finding | On a path to a crown jewel? | Steps from internet | What to do | |---|---|---|---| | Live cloud key in a config file | Yes, the key can read the payments database | 2 | Rotate today, then remove from the file | | Debug route answers in production | Yes, it exposes that config | 1 | Remove the route | | Outdated package on an internal batch job | No | Not reachable | Schedule the upgrade | | Container runs as root in CI | Not directly | 3 or more | Fix in the next sprint | | Missing security header on the marketing site | No | 1, but no data behind it | Fix when convenient |
A worked example
This example is made up. Amara runs engineering at a 12-person payments startup in Nairobi with an office in Lisbon. A scan returns about 400 results. Sorted by printed severity, the top of the list is a batch of critical CVEs in a reporting tool that runs on an internal schedule and reads no customer data.
She draws the paths. The API at api.example.com has a /debug route that answers in production. That route prints environment variables, including a cloud key for the payouts service. The key still works and can read the payments database. Three steps from anyone on the internet to card tokens.
Each finding on that route looked moderate on its own. Together they are the most urgent thing in the pile. She rotates the key and removes the route that afternoon. The next scan confirms both are gone, and she keeps that scan as evidence. The critical CVEs go into next sprint.
Mistakes we see
- Sorting by printed severity. It ranks bugs, not risk to you.
- Forgetting old domains. A forgotten staging site is an entry point with nobody watching it.
- Treating a leaked key as fixed once it is deleted from the repo. If the key still works, it is still a link. Rotate it.
- Fixing every step of a path at once. Cut the cheapest link first, then come back for the rest.
- Closing findings without re-checking. "We fixed it" is weaker than the scan from the next morning.
What to measure
- Open paths to crown jewels. The number that should fall first and stay near zero.
- Shortest path length. If your shortest route is still one or two steps, that is the week's work.
- Time from a path appearing to it being closed.
- Findings closed without a re-scan. This should be zero.
What we built
Sightline is the path method above, run on every scan.
- Connect. You give read-only access to repositories and list your live URLs. At setup you tell us your cloud provider and where sensitive data lives, and those stores are marked as crown jewels.
- Scan. Fourteen open-source scanners run in one sandboxed pass across code, secrets, dependencies, infrastructure files, Dockerfiles and live URLs. Code exists only in a temporary checkout and is deleted after the scan.
- Clean up. Duplicates across tools merge. Known-safe results drop out with the reason kept on record.
- Rank. Every finding is placed on one asset graph and scored by how short its path to a crown jewel is.
- Explain. Each finding gets a plain-English explanation, a CVSS score and a drafted fix. The model that reads your code has no tools and no network access. A separate step that uses tools never sees the code.
- Approve. A person approves each finding before it reaches you, and nothing is applied to your code without approval.
- Re-check. After a fix to code, a secret or a dependency, Sightline re-runs the scanner family that found it and confirms it is gone. If it cannot reach the repo, the result is marked inconclusive, never fixed.
What it does not do yet
Sightline does not connect to AWS, Azure or Google Cloud accounts to read running resources. Cloud resources come from your infrastructure code and what you tell us at setup, so anything created by hand in a console will not appear unless you add it. There is no runtime agent on your servers. The scan sandbox is a hardened process today, not a separate virtual machine. If your risk is mostly in a large running cloud estate, read our comparison with Wiz and Orca first.
How to judge any attack-path tool
Ask it to show a path on your own systems, not a sample. Then check:
- Does the path end at a data store you actually care about?
- Can you see why each link exists, such as which key, which route, which permission?
- Does it tell you which single fix breaks the path?
- Does a fix get confirmed by a re-scan, with a record kept?
- Does it use public exploit signals such as KEV and EPSS alongside your own layout?
Talk to us with one repo and one domain, and we will show you the graph Sightline builds. More detail is in how Sightline works.
Common questions
What is an attack path?
It is a chain of steps from a point an attacker can reach, usually the internet, to sensitive data. Each step is a weakness or a permission that lets them move further. Ranking by path tells you which findings matter together.
Is CVSS enough to prioritise?
FIRST says the CVSS base score measures severity and should be supplemented with analysis of your environment. Use it alongside exploit signals such as CISA's KEV catalog and EPSS, and your own map of where data lives.
What is EPSS?
EPSS is a model run by FIRST that estimates the probability that a published CVE will be exploited in the wild in the next 30 days. Scores are updated daily and the data is free. It tells you about the vulnerability, not about your setup.
Do I need a tool to do this?
No. A whiteboard, a list of data stores and a list of domains gets you a long way. A tool helps when you want it redone on every scan and kept current.
Does Sightline read my cloud accounts?
Not yet. It scans your repositories, infrastructure files and live URLs, and builds the rest of the graph from what you tell us at setup.