Web Search Plus v2.4.0

Web Search Plus v2.4.0

This release focuses on runtime speed, research-mode latency, and cleaner provider-key discovery for Hermes profile installs.

Credits / PR attribution

• #50 by @robbyczgw-cla — in-process web_search_plus/web_extract_plus execution and parallel research mode.
• #46 by @robbyczgw-cla — Hermes profile .env loading for provider keys.
• #49 by @wysie — plugin update instructions in the README.

Changed

• web_search_plus and web_extract_plus now run in-process by default, avoiding per-call subprocess startup overhead while keeping the subprocess path as an automatic fallback and via WSP_FORCE_SUBPROCESS=1. (#50)
• Research mode now queries providers concurrently instead of sequentially, so wall-clock time tracks the slowest launched provider rather than the sum of provider calls. (#50)
• Standalone search.py, config helpers, and setup status now load provider keys from the active Hermes profile .env as well as plugin-local legacy .env files. (#46)
• README now documents how to update an installed plugin and when to reload/reset Hermes. (#49, thanks @wysie)

Hardening

• Added bounded retry jitter to avoid synchronized retry bursts. (#50)
• Guarded provider-health updates with a lock for safer concurrent in-process calls. (#50)

Validation

• Local release gate: pytest tests/ -q → 204 passed; ruff check . → passed; compile smoke and CLI search/extract smokes passed.
• Deep local post-release smoke: doctor edge matrix 15/15 passed; doctor CLI JSON/text/live matrix passed; golden eval smoke produced 10 rows, 0 errors, 0 flagged; fresh clone focused tests passed.
• GitHub Actions: Python 3.8 and 3.11 CI passed on the release commit.

Performance note: local microbenchmarks showed a large reduction in wrapper overhead and roughly 4× lower research-mode wall time in the benchmark scenario, but real-world latency remains provider/network-bound.