Browser agents need a hard boundary.

Review snapshot

Commit

6e1977daa0f67c9de0bc0e16aaec8b5833eeb8e0

Python files

274 bounded · 0 syntax errors

Engine findings

408 total · 244 info · 161 low · 3 medium

Coverage note

273 generic parser limitations — that is, 273 of the 274 files matched no framework-specific adapter matched and were analysed by the generic heuristic adapter instead. All 274 files parsed, with 0 syntax errors; “uncovered” means unrecognised by a framework adapter, not unread, which is why 118 of them still carry findings. A generic-adapter result is weaker evidence than a framework-adapter one, a parser limitation is not a finding, and a file that was not covered by a framework adapter was not concluded to be safe.

Scan scope id

tree_sha 419ab9358e627eae — this is a scan-scope and tree-shape identifier, not a content hash and not a cryptographic digest: it is not collision-resistant, and AxiomGate's own verification script demonstrates a collision between differing trees of equal file sizes. The upstream git commit above is what binds the content.

Published artifact digest

SHA-256 of the published findings.json artifact: fc2a27944c3df9390e2968d5406af51f7f4ffd930cd6a9db3d8388ed01d975b2 — this identifies the published evidence file itself. It is a different kind of identifier from the upstream git commit (which binds the analysed source), from the scanner tree_sha above (a scan-scope identifier, not a content hash), from the demo fixture digest, and from the formal artifact hashes. Verify it with sha256sum findings.json.

Raw evidence

findings.json — every automated finding with its rule id, rule version, source location, confidence and explicit “cannot claim” statement, plus the exact analyzer configuration used. Automated static output only; the manual review section below is separate and is not analyzer output.

Review themes

5 architectural hardening gaps

Revalidated

2026-09-18 · analyzer checkpoint 21419c03

Publication

Local only