gstack/browse/src
Garry Tan 07edc70df1
feat(security): Bun-native inference research skeleton + design doc
Ships the research skeleton for the P3 "5ms Bun-native classifier" TODO.
Honest scope: tokenizer + API surface + benchmark harness + roadmap doc.
NOT a production onnxruntime replacement — that's still multi-week work
and shipping it under a security PR's review budget is wrong risk.

browse/src/security-bunnative.ts:
  * Pure-TS WordPiece tokenizer reading HF tokenizer.json directly —
    produces the same input_ids sequence as transformers.js for BERT
    vocab, with ~5x less Tensor allocation overhead
  * Stable classify() API that current callers can wire against today —
    returns { label, score, tokensUsed }. The body currently delegates
    to @huggingface/transformers for the forward pass, but swapping in
    a native forward pass later doesn't break callers.
  * Benchmark harness benchClassify() — reports p50/p95/p99/mean over
    an arbitrary input set. Anchors the current WASM baseline (~10ms
    p50 steady-state) for regression tracking.

docs/designs/BUN_NATIVE_INFERENCE.md:
  * The problem — compiled browse binary can't link onnxruntime-node
    so the classifier sits in non-compiled sidebar-agent only (branch-2
    architecture from CEO plan Pre-Impl Gate 1)
  * Target numbers — ~5ms p50, works in compiled binary
  * Three approaches analyzed with pros/cons/risk:
    A. Pure-TS SIMD — ruled out (can't beat WASM at matmul)
    B. Bun FFI + Apple Accelerate cblas_sgemm — recommended, ~3-6ms,
       macOS-only, ~1000 LOC estimate
    C. Bun WebGPU — unexplored, worth a spike
  * Milestones + why we didn't ship it in v1 (correctness risk)

Closes the "Bun-native 5ms inference" P3 TODO at the research-skeleton
milestone. Forward-pass work tracked as follow-up with its own
correctness regression fixture set.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-20 05:02:59 +08:00
..
activity.ts feat: content security — 4-layer prompt injection defense for pair-agent (#815) 2026-04-06 14:41:06 -07:00
audit.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
browser-manager.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
buffers.ts feat: Phase 3.5 — cookie import, QA testing, team retro (v0.3.1) (#29) 2026-03-13 00:31:41 -07:00
bun-polyfill.cjs fix: Windows support — Node.js server fallback for Playwright (#255) 2026-03-20 12:22:11 -07:00
cdp-inspector.ts refactor: AI slop reduction with cross-model quality review (v0.16.3.0) (#941) 2026-04-10 17:13:15 -10:00
cli.ts fix: headed browser auto-shutdown + disconnect cleanup (v0.18.1.0) (#1025) 2026-04-16 15:39:44 -07:00
commands.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
config.ts fix: security wave 3 — 12 fixes, 7 contributors (v0.16.4.0) (#988) 2026-04-13 07:49:37 -10:00
content-security.ts refactor: AI slop reduction with cross-model quality review (v0.16.3.0) (#941) 2026-04-10 17:13:15 -10:00
cookie-import-browser.ts community wave: 6 PRs + hardening (v0.18.1.0) (#1028) 2026-04-17 00:45:13 -07:00
cookie-picker-routes.ts community wave: 6 PRs + hardening (v0.18.1.0) (#1028) 2026-04-17 00:45:13 -07:00
cookie-picker-ui.ts fix: cookie picker auth token leak (v0.15.17.0) (#904) 2026-04-08 10:10:13 -07:00
error-handling.ts refactor: AI slop reduction with cross-model quality review (v0.16.3.0) (#941) 2026-04-10 17:13:15 -10:00
find-browse.ts feat: multi-agent support — gstack works on Codex, Gemini CLI, and Cursor (v0.9.0) (#226) 2026-03-19 18:20:50 -07:00
media-extract.ts feat: browser data platform for AI agents (v0.16.0.0) (#907) 2026-04-08 00:41:55 -07:00
meta-commands.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
network-capture.ts feat: browser data platform for AI agents (v0.16.0.0) (#907) 2026-04-08 00:41:55 -07:00
path-security.ts fix: security wave 3 — 12 fixes, 7 contributors (v0.16.4.0) (#988) 2026-04-13 07:49:37 -10:00
platform.ts fix: Windows support — Node.js server fallback for Playwright (#255) 2026-03-20 12:22:11 -07:00
read-commands.ts fix: security wave 3 — 12 fixes, 7 contributors (v0.16.4.0) (#988) 2026-04-13 07:49:37 -10:00
security-bunnative.ts feat(security): Bun-native inference research skeleton + design doc 2026-04-20 05:02:59 +08:00
security-classifier.ts feat(security): DeBERTa-v3 ensemble classifier (opt-in) 2026-04-20 04:55:23 +08:00
security.ts feat(security): 3-way ensemble verdict combiner with deberta_content layer 2026-04-20 04:55:23 +08:00
server.ts feat(security): shield icon continuous polling via /sidebar-chat 2026-04-20 04:39:18 +08:00
sidebar-agent.ts feat(security): DeBERTa-v3 ensemble classifier (opt-in) 2026-04-20 04:55:23 +08:00
sidebar-utils.ts fix: sidebar agent uses real tab URL instead of stale Playwright URL (v0.12.6.0) (#544) 2026-03-26 22:07:03 -06:00
snapshot.ts feat: UX behavioral foundations + ux-audit command (v0.17.0.0) (#1000) 2026-04-14 07:47:11 -10:00
tab-session.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
token-registry.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
url-validation.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00
welcome.html feat: GStack Browser — double-click AI browser with anti-bot stealth (#695) 2026-04-04 10:17:05 -07:00
write-commands.ts feat(browse): Puppeteer parity — load-html, screenshot --selector, viewport --scale, file:// (v1.1.0.0) (#1062) 2026-04-18 23:25:33 +08:00