gstack/browse
Garry Tan b4e49d080d
feat(security): 3-way ensemble verdict combiner with deberta_content layer
Updates combineVerdict to support a third ML signal layer (deberta_content)
for opt-in DeBERTa-v3 ensemble. Rule becomes:

  * Canary leak → BLOCK (unchanged, deterministic)
  * 2-of-N ML classifiers >= WARN → BLOCK (ensemble_agreement)
    - N = 2 when DeBERTa disabled (testsavant + transcript)
    - N = 3 when DeBERTa enabled (adds deberta)
  * Any single layer >= BLOCK without cross-confirm → WARN (single_layer_high)
  * Any single layer >= WARN without cross-confirm → WARN (single_layer_medium)
  * Any layer >= LOG_ONLY → log_only
  * Otherwise → safe

Backward compatible: when DeBERTa signal has confidence 0 (meta.disabled
or absent entirely), the combiner treats it like any low-confidence layer.
Existing 2-of-2 ensemble path still fires for testsavant + transcript.

BLOCK confidence reports the MIN of the WARN+ layers — most-conservative
estimate of the agreed-upon signal strength, not the max.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-20 04:55:23 +08:00
..
bin feat: multi-agent support — gstack works on Codex, Gemini CLI, and Cursor (v0.9.0) (#226) 2026-03-19 18:20:50 -07:00
scripts fix: ngrok Windows build + close CI error-swallowing gap (v0.18.0.1) (#1024) 2026-04-16 13:49:04 -07:00
src feat(security): 3-way ensemble verdict combiner with deberta_content layer 2026-04-20 04:55:23 +08:00
test test(security): add BrowseSafe-Bench smoke harness (v1 baseline) 2026-04-20 04:50:53 +08:00
PLAN-snapshot-dropdown-interactive.md fix: snapshot -i auto-detects dropdown/popover interactive elements (#845) 2026-04-05 22:57:45 -07:00
SKILL.md feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
SKILL.md.tmpl feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00