On a pull request it asks that question before the merge: the check goes red only when this change opens a route, and the fix comes back as its own pull request. Open source (Apache 2.0), runs on your infrastructure, collects no telemetry.
Twelve seconds of make demo: what is exploitable now → the ranked routes → one route's kill
chain and the fix it generates → whether the scores can be trusted. Sample scanner output and
seeded verdicts, not a real environment.
- See it running - the same dashboard, published read-only. Nothing to install.
- Read the documentation - the manual, one page per subject, with search.
- Check your own AWS account - one read-only command, no deployment.
- Put it on your pull requests - ten lines of YAML.
A score here is what the model concludes from the evidence it was given, not a measured frequency: nothing has been calibrated against field data yet, and the engine says so itself rather than rounding up. What is measured, and what is not.
Check your own account in 30 seconds
No deployment, no Docker, nothing ingested. One static binary asks AWS's own policy evaluator which of your roles can reach administrator - applying the service control policies, permission boundaries and condition keys that a policy reader on its own does not see:
# macOS (Apple silicon); swap darwin_arm64 for linux_amd64, linux_arm64 or darwin_amd64
curl -sSL https://github.com/luiacuaniello/perspectivegraph/releases/latest/download/perspectivegraph_darwin_arm64.tar.gz | tar xz
./perspectivegraph redteam -roles -region eu-west-1
It is read-only and free: each check is one iam:SimulatePrincipalPolicy dry run, which
creates nothing and costs nothing, and the only permissions it needs are that and
iam:ListRoles, both inside SecurityAudit. Windows builds are on the
releases page; every binary
is signed with cosign and carries SLSA provenance, and two commands
verify both before you run anything.
Add -compare and it also runs the engine over the same account, exiting non-zero where the
two disagree. That is how the engine's
first real false positive
was found, and how it stays fixed. If it disagrees on yours,
report it: no report is more useful. From here, how to evaluate this walks to a
verdict on your own estate in stages that each end in an answer.
See the whole engine in 90 seconds
make demo
Pulls the published, cosign-signed images, feeds them sample Trivy / Semgrep / Custodian /
Falco / Kubernetes / IAM / SSO output, and prints the top attack path with its generated fix.
Dashboard on http://localhost:3000, make down to tear it down. It needs Docker, jq and
curl, compiles nothing, and takes about 23 seconds from an empty image cache; make demo-build
is the same demo built from your working tree.
On Kubernetes, the chart is an official package on Artifact Hub:
helm install perspectivegraph oci://ghcr.io/luiacuaniello/charts/perspectivegraph \
--version 1.39.1 # x-release-please-version
On one machine - a VM with Docker and a DNS name - one command sets up TLS, generated credentials, an authenticated bus and daily backups, and starts it: production on one machine.
make prod-init DOMAIN=perspectivegraph.example.com
The chart and the three images it runs (ghcr.io/luiacuaniello/perspectivegraph, -dashboard
and -postgres) are signed with cosign keyless and carry an SPDX SBOM and SLSA provenance:
verify them rather than taking the supply chain on trust.
The dashboard opens on the decision, not the inventory: what is being exploited now, the fewest changes that remove the most risk, and how far the numbers can be trusted. Routes are ranked by triage priority - what the route reaches, whether runtime confirmed it, how exposed the entry is - so a lower-scoring route can outrank a higher-scoring one.
 |  |
| Why the route is P1, one probability with its range, and every hop with what it lets the attacker do and where its probability came from. | Whether the engine's own scores held up against recorded outcomes - and whether there are enough of them to say. |
The screenshots are make demo with seeded verdicts: make seed-validation records
synthetic outcomes on part of the routes, the way a BAS run tests part of an estate, which is why
the calibration panel has something to show - and why it still says not enough outcomes yet:
the handful it records is far short of the 30 a verdict needs. The counts in the pictures are
those of the release they were taken on; the sample data has grown since, so yours differ by a
route or two. A fresh install and the public demo report
insufficient data instead, until real outcomes exist. The public demo runs on one free VM, so
treat it as best-effort.
Why?
A scanner reports that a container carries a critical CVE. It cannot report that the container sits behind an internet-facing load balancer, runs with a role that reads the production database, and is therefore the one finding out of ten thousand worth fixing this week. That needs the other tools' output in the same graph, which is what this builds. A developer gets a check that goes red only when their change opens a real route; a security team gets a short ranked list of attack paths instead of a flat list of findings.
Block the pull request that opens the path
No deployment required. The runner reads your estate read-only, ingests this pull request's scan, and answers in-process with the same engine:
- uses: luiacuaniello/perspectivegraph@v1
with:
mode: local
aws-region: eu-west-1 # read-only; give the job an OIDC role with SecurityAudit
report: trivy.json
The check goes red when this change opens a route to a sensitive asset - or makes one likelier - not when it adds a critical CVE: a critical on a host nothing routes to does not fail the build, and a medium on a container that now reaches the production database does. Routes that were there before the change do not count, even when they run through what it touches: the scan is applied to a copy of the estate and compared with it, and nothing is written. It also has a third outcome, because a pipeline whose scan never arrived must not get the same green tick as one that is clean:
A Terraform plan goes through the same gate before anything is deployed: source: terraform
with the output of terraform show -json, judged on its own or against the live estate. A
security group rule, a policy attachment or a bucket policy that opens a route fails the check
in the pull request that writes it - see Terraform plans.
Outside GitHub Actions it is one command, and it installs as a Trivy plugin too:
perspectivegraph gate -local -aws-region eu-west-1 -report trivy.json -slug owner/name -sha "$COMMIT_SHA"
trivy plugin install github.com/luiacuaniello/perspectivegraph
trivy image -f json myapp:pr-42 | trivy perspectivegraph gate -local -aws-region eu-west-1 -report -
[!WARNING] Fork pull requests get no secrets, so the gate fails closed as
unknownon them. Do not work around it withpull_request_target: that runs your secrets - in local mode, cloud credentials - against the contributor's code.On a public repository, a blocked check prints the route (real asset names, the CVE, the sensitive asset) into a public job log. Use
soft-failand post the detail somewhere private.
The manual covers the rest:
pointing the action at a deployed engine, gating rendered manifests, a pre-collected estate,
rolling the gate out, and every input in action.yml.
Let an agent query it
A language model cannot enumerate thousands of edges reliably or run Dijkstra, and asked for "the attack paths in my account" it will invent plausible ones. So the engine speaks MCP: the agent asks, and reasons over answers it could not have made up.
perspectivegraph mcp --api http://localhost:8080 # or https://demo.a3thinker.it, no credential needed
Eight tools, every one read-only and declared so on the wire. The one worth the integration
is simulate_fix: it re-runs the simulation with the given edges cut and reports what actually
changes. The server is in the official MCP Registry
and on Glama; the tools and the
client configuration are in the manual.
Project status & maturity
The engine and its public API are complete, documented and tested. The scores are not yet calibrated. Nobody has run this over a real estate, tested the paths it surfaced and fed the verdicts back: the machinery for that loop is built and tested, the loop is not closed. So read a score as what the model believes and how sure it says it is, not as a measured frequency. Use it to find and cut routes; don't put its percentage in front of a board. If you run it over yours, tell us how it went. Positioning spells out what is and isn't claimed.
What is measured today, as of v1.39.1.
make bench-cloudgoat grades the engine in CI on four
CloudGoat-shaped scenarios:
Precision and recall are 1.00 on all four: four known shapes, two of them negative controls, so a
regression gate rather than a measure of accuracy on your estate. On real AWS,
make reachability-lab-aws checks exposure the same way for free, and make redteam-aws grades
the engine's escalation claims against AWS's own policy evaluator. That grading has already
caught one false positive, a permissions boundary the engine ignored; it is fixed, and
make boundary-lab-aws fails whenever the engine and AWS disagree. make entrypoints-lab-aws
puts bucket policies and function URLs to AWS the same way; its first run caught four errors,
also fixed. make public-access-lab-aws does it, for free, for Block Public Access set on the
whole account, and make terraform-lab-aws applies the Terraform plans the merge gate judged and
asks the internet whether each one opened a way in. Each of these labs writes what it found
into the build, and the dashboard's Accuracy page shows it under Checked against AWS: every
question, AWS's answer, the engine's, and the version it was asked of.
- Clouds. AWS is live and verified against a real account; reading several accounts in one pass is exercised on fixtures, not yet against real ones. Azure is fixtures only, and there is no connector for GCP or for OpenStack: the roadmap says what is planned for each.
- Interface. The GraphQL schema is frozen and drift-guarded, so a breaking change comes with a major version, never in a patch: API stability.
- Deployment. The backend is distroless and non-root on a read-only root filesystem, the
images are pinned by digest, and under Helm every workload meets the
restrictedPod Security Standard, asserted in CI. The compose defaults are open on purpose;PG_ENV=productionrefuses to start unless API and ingest are authenticated. A real rollout needs more (TLS, backups,TRUSTED_PROXY_CIDRSbehind a proxy, a database you can fail over): on one machinemake prod-initdoes the first three, and CI restores one of its backups on every change; the operations runbook lists the rest. - Support. The newest release only, with a clock on security fixes (Critical 7 days, High 30): SUPPORT.md.
- Telemetry. None. Out of the box it opens no outbound connection; GitHub, the AI assistant and the KEV/EPSS feeds each stay off until you configure them.
- Scope. The reachable attack-path question, in the developer workflow. It is not a scanner, a CNAPP or a compliance product, and replaces none of them.
- How it is written. By a human working with Claude (Anthropic): the design decisions and
what ships are the maintainer's, and much of the implementation and its tests came out of that
collaboration. Check it rather than trust it:
make test,make bench-cloudgoat,govulncheck ./..., and CONTRIBUTING.
Documentation
- Concepts in plain words - route, score, interval, priority, outcome, calibration: the words the dashboard uses
- Connect your own infrastructure - your AWS accounts and Kubernetes clusters, from a laptop, one machine, EKS or any cluster
- Manual - architecture, scoring, every integration, deployment, and the runbook for your own environment
- Evaluation - trying it on your own estate, in stages that each end in an answer
- Positioning - what is claimed, what is not, and how to check
- Operations · Threat model · Scale · API stability
- Support · Upgrading · Roadmap · Security
- Contributing · Governance · Maintainers · Adopters
License
Apache License 2.0. Licensing is declared for every file, in the form REUSE defines, and checked in CI: REUSE.toml.