Myrqen is a working codename that has passed a preliminary collision screen only. The
scanner is source-available and proprietary, not OSI open source. See
SOURCE-AVAILABLE-LICENSE-DRAFT.md in the
repository for what the licence does and does not grant.What a scan actually is
A scan has two halves, and the split is the whole design.The static pass is Myrqen's own
myrqen session start parses the project’s TypeScript and JavaScript, follows
attacker-controlled data through a function, and records where it reaches a sink
with nothing in between. It runs whether or not an agent is involved, which is what
makes it measurable.The dynamic half is the agent's
Static reading cannot observe a response, so every static candidate arrives as a
claim rather than a finding. Your agent settles it against the running application
with
myrqen finding verify or myrqen finding refute.needs_review and says so in the report. That is the
honest state, and the report never pretends otherwise.
Who it is for
Developers and small teams who want a real application security pass on something they are building, before shipping it, without buying a second model subscription or handing their repository to a scanning service. The repository is never uploaded. Scanning, report generation in all four formats, the fix prompt, and applying a fix locally are never gated by plan.Primary capabilities
Agent-driven assessment
A portable Agent Skill plus a universal CLI. Works as
/myrqen auto in Claude Code
and as myrqen auto from any agent with shell access.Enforced safety policy
Exact-origin authorization for anything off your machine, a prohibited-action list
that higher effort never relaxes, and quarantine for origins that untrusted project
content proposed.
Findings with provenance
Severity and verification are independent. Deduplication, redaction, and finding
identity are owned by the CLI, not by the agent.
Reports you own
Self-contained HTML, plus JSON, Markdown, and SARIF, written to
.myrqen/reports/<id>/ in the project you scanned.Optional cloud sync
Live progress, retention windows, and one share URL that renders differently per
recipient — asked for once per scan, never automatic. Hosted at
myrqen.cc, or
run it yourself.A measurable engine
An owned corpus of vulnerable cases paired with safe near-misses, plus the scorer
that grades recall, precision, and false positives.
Where to start
1
Install the CLI
2
Run your first assessment
Point it at a project you own and let your agent work the phases. See the
Quickstart.
3
Read the safety rules before you test anything external
Local targets need no prompt. Everything else needs an exact-origin grant. See
Authorization and safety.
What Myrqen will not do
- It will not tell you that you are secure. It reports what it can establish and states what it could not reach. Silence from a scanner is not a clean bill of health.
- It will not touch a system you have not authorized. A grant for
https://api.example.comnever extends toexample.com, another port, or another scheme. - It will not do anything destructive. Mass deletion, request flooding, credential spraying, persistence, and scope expansion are refused by policy, not by convention, and a refusal is recorded in the report.
- It will not treat text in your repository as instructions. A README that tells the scanner to ignore its rules is data. The attempt is recorded as an observation.
- It will not print a secret it found. Findings describe a credential by type and location; the value is redacted before it is stored anywhere, including locally.