v0.2.0

Highlights

• Added read tool support for one-shot baseline reads via bypass_cache?: boolean.
  • When bypass_cache: true, readcache optimization is skipped for that call and baseline output is returned.
  • Useful when the model needs exact baseline content without unchanged/diff compaction.
• Tightened full-file diff usefulness thresholds to reduce noisy near-full-file patches:
  • MAX_DIFF_TO_BASE_RATIO: 1.0 -> 0.9
  • MAX_DIFF_TO_BASE_LINE_RATIO: 1.0 -> 0.85
• Improved fallback behavior for high-churn edits: large/unhelpful patches now fall back to baseline content sooner (mode=baseline_fallback).
• Added README documentation for diff gating behavior, thresholds, and rationale.

Why this release

This release improves read output quality for LLM consumption by emitting diffs only when they are clearly useful, while also providing an explicit one-shot bypass path when baseline output is required.

Full Changelog

v0.1.1...v0.2.0

v0.1.1

Fixed

• readcache_refresh range invalidation now reliably forces the next read of that exact range to return a baseline payload.
• Prevented incorrect fallback from invalidated range scope to full-scope trust.
• Range scope optimization is restored after the range is re-anchored with a baseline read.

Tests

• Added regression coverage for:
  • range refresh blocking full fallback
  • range refresh -> full read -> next range read baseline behavior
  • replay/runtime blocker behavior

Other

• Updated documentation path references.
• Bumped package version to 0.1.1.

v0.1.0

Full Changelog: https://github.com/Gurpartap/pi-readcache/commits/v0.1.0