Welcome — and v0.5.0 is out

[quolu]

Welcome — and v0.5.0 is out.

Throughline cuts ~90% of Claude Code's context usage while keeping nearly all the memory, by separating conversation by content type, not by recency:

• Text (your words + Claude's replies) stays in-context as L1/L2
• Tool I/O (file reads, Bash output, grep) retires to SQLite as L3, retrievable on demand via /sc-detail HH:MM:SS

That means /clear no longer wipes your work — the new session resumes mid-thought.

What's new in v0.5.0

This release closes out the v0.5 transcript-injection investigation and locks in path C (resume-context.mjs v2.1 header + 現在地 anchor) as the plugin-scope completion form.

• Strengthened the Claude /clear resume context header with two new short-message handling rules so the cleared-me side stops misreading follow-up shorts ("OK", "進めて", "次") as fresh requests
• Two alternative injection routes (transcript JSONL append, hookSpecificOutput.initialUserMessage) spiked end-to-end against real Claude Code v2.1.145 — both confirmed dead, locking path C as the plugin-scope ceiling
• Full investigation log in docs/THROUGHLINE_TRANSCRIPT_INJECTION_PLAN.md and grounding spec extracts in docs/RAG/

Full release notes: https://github.com/kitepon-rgb/Throughline/releases/tag/v0.5.0

Where to talk to us

• Q&A — got a setup question, a hook not firing, tokens not showing in the monitor? Drop it here.
• Ideas — what context-management primitive would you like next? Codex parity gaps, new monitor layouts, etc.
• Show and tell — paste a screenshot of your monitor, share a /clear resume that actually worked across a real refactor.
• Announcements — release notes will be cross-posted here.

Quick start (in case you just landed)

npm install -g throughline
throughline install     # registers hooks/skills, provisions the VS Code monitor task

Open any Claude Code session, work for 50 turns, type /clear. The new session resumes from where you left off instead of from zero.

— Maintainer