Verifies every finding by execution
A finding is marked verified only when its reproduction command succeeds in a ephemeral microVM. The environment issues the verdict, not the model.
orca is an authorization-gated research agent. it finds the vuln, reproduces it in a clean microvm, and writes the report — inside the scope you declare.
The run from a declared surface to a written report. Nothing reaches the report that has not been reproduced.
subfinder, httpx, naabu, katana and wayback wrap the recon; daemon mode adds crt.sh and NVD feeds.
Leads become ledger entries, each carrying the evidence it would take to prove it.
Reproduction in a ephemeral microVM, diff_probe and race_probe, and the OOB listener for blind vulns.
Verified primitives assemble into the chain graph: entry, primitives, goal.
report.md covers verified paths only; every tool call left a receipt in .cyber/.
Modes one engagement or a daemon; OOB listener, headless browser, fuzzer with triage.
A finding is marked verified only when its reproduction command succeeds in a ephemeral microVM. The environment issues the verdict, not the model.
Targets and the authorization artifact are declared in cyber.toml; every command gets a gate verdict before it runs.
A lead enters the ledger with the evidence it needs and stays there until it is verified or killed, with the reason kept.
report.md carries reproduced primitives, the command that proved them and the audit receipts. Nothing unverified gets in.
A finding enters the report only after its reproduction command has run once, in a ephemeral microVM.
Targets and the authorization artifact live in cyber.toml. Every command gets a gate verdict first. Out of scope is refused before it runs and logged to .cyber/.
[engagement] id = "lab-q1" operator = "lazarevka" authorization = "artifact a-118, signed" scope = ["lazarevka.xyz", "*.lazarevka.xyz"] goal = "reach the admin panel" [vm] image = "alpine-baked.qcow2"
An authorization-gated vulnerability research agent for security teams on authorized engagements. It maps a declared attack surface, tracks exploit hypotheses in a ledger, and assembles verified primitives into a chain graph.
No. A scanner reports possible matches; orca marks a finding verified only after its reproduction command succeeds in a ephemeral microVM. It works only inside declared scope and does not invent new bug classes.
The engagement is declared in cyber.toml: targets and an authorization artifact. Every command passes a gate verdict before it runs; out-of-scope commands are refused before they execute, and every call writes a receipt into .cyber/.
Security teams running authorized engagements who need findings they can defend. Access is proprietary and per engagement — request access to start one.