v0.7.16

[0.7.16] - 2026-04-09

Fixed

• Bounded review draft chunking now comes from a centralized provider budget policy instead of provider-name heuristics, with explicit execution metadata for strategy, estimates, envelopes, and chunk reasons
• Review-mode markdown file context is now sliced into heading-aware evidence packs with anchored source metadata, instead of relying only on first-byte truncation
• Downstream critique and synthesis phases now consume normalized chunk findings and quoted evidence excerpts instead of silently replaying the full raw file bundle after chunked drafts
• OpenRouter bounded critic --mode review runs now skip impossible over-budget critique calls and return a conservative reviewer fallback object when repeated synthesis responses fail JSON validation

Changed

• Added large-review observability to the execution plan, including truncation warnings, slicing metadata, per-phase prompt sizing, and bounded prompt-compaction decisions
• Updated the OpenRouter quickstart, packaged skill metadata, and Claude plugin version for the 0.7.16 release