CyberSecurityUP 5af21f13f0 feat(harness): add LLM dedup, budget warnings, agent nudge, image budget, and run resume ...

- LLM-based finding dedup: fast-path deterministic (title+endpoint, CWE+endpoint) + LLM semantic judge for fuzzy cases
- Progressive budget warnings at 70/85/95% with stop-new-agents gate at 95%
- Token tracking on ModelPool (record_tokens on every completion, set_token_limit from budget config)
- Agent lifecycle nudge: retry failed agents up to 3x with progressively simplified prompts
- Image budget enforcement: cap base64 screenshots in prompts, strip-all for recovery
- Run snapshot/resume: persist RunSnapshot to workdir, skip already-completed agents on resume
- 21 unit tests for all new features (500 total pass)

2026-10-18 19:52:10 -03:00

..

agents_md/mobile

feat(mobile): binary/APK/IPA testing mode + 12 RE skills — v4.2.0

2026-09-20 21:08:14 -03:00

app

release: v4.3.3 — per-finding attack path, LLM config modal, severity badges

2026-10-16 17:26:42 -03:00

crates/harness

feat(harness): add LLM dedup, budget warnings, agent nudge, image budget, and run resume

2026-10-18 19:52:10 -03:00

templates

fix(web+report): treat needs-review findings as validated findings

2026-10-18 16:54:49 -03:00

tools

feat(decision): pluggable System One backend — TypeSafe (hosted) or Laya (local)

2026-09-20 19:27:14 -03:00

.gitignore

chore: remove test artifacts, ignore scan debris (#35)

2026-07-31 15:39:00 -03:00

Cargo.lock

feat(web+report): Wiz-style attack path nodes, inline finding details, PDF formatting

2026-10-16 17:38:02 -03:00

Cargo.toml

release: v4.3.3 — per-finding attack path, LLM config modal, severity badges

2026-10-16 17:26:42 -03:00

creds.example.yaml

…