Summary

• add Pipeline::evaluate_through_l2 with structured L1, L2, effective-decision, stage-status, and escalation-cause output
• preserve unresolved escalation without invoking L3 or applying fail_closed and speculative-L3 resolution
• expose the through-L2 path through the Rust SDK and an asynchronous TypeScript N-API worker task
• consume Observer argv_truncated and argv_incomplete signals so incomplete command evidence cannot become an ordinary allow decision
• keep explicit dangerous prefixes blockable while stopping incomplete ambiguous evidence at L1
• version the TypeScript SDK as 0.2.0 and update SDK documentation, lockfiles, and test coverage

Why

AnySentry needs to execute L1/L2 in a fast-judge worker and hand unresolved L2 escalations to an independent, durable L3 worker.

The existing evaluate path may invoke L3 synchronously or resolve an outstanding escalation through fail_closed, which hides the intermediate state required for asynchronous dispatch and can block the AnySentry API and Dashboard while L3 is running.

Observer now reports command-evidence quality explicitly. Sentry must preserve that uncertainty instead of allowing an event based on a benign-looking but incomplete argv prefix.

Validation

• cargo fmt --all -- --check
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test
  • 67 unit tests passed
  • 12 integration tests passed
• npm run build
• npm test
  • 11 TypeScript SDK tests passed
• git diff --check
• AnySentry API and web builds passed against the locally packed TypeScript SDK
• AnySentry Observer-ingest integration verification passed