v0.26.7

Fix reddit-fetch skill: use old.reddit.com, jq parsing, rate limit guidance

v0.26.6

• Add system prompt patches for v2.1.81

v0.26.5

Add YouTube transcript tip to Cmd+A section

v0.26.4

• Add system prompt patches for v2.1.79
• Show 1M instead of 1000k in status line context display

v0.26.3

• Update voice mode link and add personal note
• Add system prompt patches for v2.1.78
• Add system prompt patches for v2.1.77
• Update VoltAgent PR #197 status to closed
• Mention paper-search plugin in Tip 27

v0.26.2

• Add native binary install steps to CLAUDE.md
• Add curl JSON API fallback to reddit-fetch skill

v0.26.1

• System prompt patches for Claude Code 2.1.71, 2.1.72, 2.1.76
• Reddit-fetch skill: add fallback for unavailable model, fix retry instruction placement
• Update dx-submissions tracker

v0.26.0

System prompt patches for Claude Code 2.1.51 through 2.1.70

• Added patches for versions 2.1.51, 2.1.63, 2.1.66, 2.1.68, 2.1.69, 2.1.70
• 2.1.70: new customRegex support in patch-cli.js for complex JS ternary expressions that differ across npm/native builds
• Added note about Claude Code built-in voice mode

All patch sets verified across npm, native Linux, and native macOS builds.

v0.25.1

fix: half-clone counts progress messages as user messages

The grep chains matched "type":"user" anywhere in the raw JSON line, including inside nested subagent progress data. This inflated the clean user message count (e.g. 53 vs actual 28), causing the cut point to land on a progress message instead of a real user turn.

Replace all grep chains with a single filter_clean_user_msgs helper that extracts the top-level type (first "type":"..." in the line). Also filters isMeta, interrupted, and tool_result in one pass.

Fixes #8.

Also fixes 4 pre-existing test failures and adds a new test for the progress message bug. 11/11 tests passing.

Plugin version: 0.14.9

v0.24.0

• Replace container folder with SafeClaw link