leon-ai/leon: technical due diligence, read from its repository

3 model weight files are stored in git itself, not in LFS, DVC or a model hub. Every clone carries them, the history keeps every version for good, and nothing records which run or data produced them.

In the last 24 months, 150 of 1,641 commits (9.1%) carry any trace of review: a merge, a pull-request reference, a Reviewed-by or Reviewed-on trailer, a Closes/Fixes reference, a maintainer's sign-off on someone else's change, or a change applied by someone other than its author. That is low but not absent. Review may happen in a workflow this history does not record; a buyer should ask how change is approved before it ships.

There is no disclosure policy, no automated dependency updates, no named code owners and no pre-commit configuration. None of these is load-bearing on its own. Their collective absence says that security posture here is whatever the current team happens to remember to do, which is not a posture that survives the team changing.

26 of 940 locked dependencies match 83 published advisories in OSV, 48 of them rated high or critical. A lockfile says which version is installed, not whether the flawed code is reachable from this product, so each needs a look. Most are fixed by an upgrade, and the advisory names the version.

The build configuration emits source maps for production, so the original source, with its comments and structure, is downloadable by every visitor. Upload them to an error tracker and keep them off the public server instead.

server/src/telemetry.ts sends to telemetry.getleon.ai, and config.sample.yml leaves reporting on as shipped. Every installation reports to the vendor unless its operator turns this off. For a buyer running it inside their own product, that is an outbound data flow to a third party: ask what is sent, how often, and whether the buyer's customers were told.