fix(api): preserve keepalive liveness through provider streams

[BunsDev]

Summary

Preserve Anthropic keepalive pings through the provider adapter so a live, quiet stream does not hit the query stall timeout.

Context

• Related user request: finish the saved dirty worktrees and retire completed work.
• Completes the remaining keepalive patch from fix/provider-stall-first-byte-timeout; follows fixes Agent stalls: model announces work, ends turn with end_turn and no tool calls — loop accepts it as done #165 and fix(query): recover announce-then-stop stalls instead of ending the turn #166.
• Scope: provider stream liveness and regression coverage. No CLI or persisted engine contract changes.

Changes

• Forward content-free Ping events to the consumer's existing deadline reset.
• Ignore pings in accumulation and transcript publication.
• Cover adapter propagation and quiet-stream completion with paused Tokio time.

Validation

Commands run from src-rust using rustup run stable (Rust 1.99) and an external target directory.

• git diff --check
• cargo fmt --all
• cargo check --workspace
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test --workspace: 2,130 passed; 7 ignored.
• Targeted: cargo test -p claurst-api keepalive_ping_reaches failed before the fix, then passed.
• Targeted: provider keepalive liveness, empty accumulator and no transcript ping regression passed.
• Not run: live authenticated provider request; deterministic stream coverage exercises the changed behavior without network credentials.

PR Readiness

• Diff is limited to the intended files.
• Generated files were not edited by hand.
• User-facing docs/help are not needed for this internal liveness fix.
• No secrets, credentials, local paths, or unrelated logs are included.
• Independent review found no production blockers; its test-timing suggestion was applied.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
docs	Ready	Preview	Oct 13, 2026 12:48pm UTC