paranoid and its /hack-me workflow are for securing your own code and pentesting your own running app, locally and with authorization. They are not built to target third-party systems, scan hosts you don't own, evade detection, or produce live malware — and the skill is written to decline when a request drifts that way. Please use it accordingly.

The example targets (examples/ledgerlite, examples/vampi) are intentionally vulnerable and meant to run only on localhost. Never deploy them anywhere reachable.

The same goes for benchmark/solutions/selftest_insecure/: every file there is a deliberately exploitable reference (SQL injection, pickle deserialization, path traversal, and so on) that exists so CI can prove the harness still detects a known bug. They are never imported by the skill or the /hack-me command, and they are not a vulnerability in this project — please don't file reports against them.

The skills and benchmark here are Markdown and dependency-free Python that your agent reads and runs locally — there's no server or hosted component. If you still find a security issue in this repository (for example, a /hack-me instruction that could be misused, or a benchmark harness that executes solution code unsafely):

• Preferred: open a GitHub Security Advisory (private disclosure).
• Or open a normal issue if it is not sensitive.

Please include what you did, what happened, and what you expected. We aim to acknowledge reports quickly and will credit reporters who want it.

benchmark/harness/run.py imports and executes the solution files you point it at. Only run it on solutions you trust or generated yourself — treat it like running any untrusted script.