State-Of-Claude-Context-0426

An openly-logged research workspace on the state of Claude's context handling as of April 2026.

Variant: general-research-workspace (provisioned by the research-space Claude Code plugin).

Question: What is the current state of context handling in Claude — context windows, prompt caching, memory, compaction, tool-result handling, and how these shape real workflows?

Q&A pattern: questions are posed, Claude gathers material and writes longform responses, paired records accumulate in questions/ + answers/, and selected pairs are periodically consolidated into PDFs or reports.

Layout

• CLAUDE.md — workspace instructions (variant: general-research-workspace)
• questions/ — one markdown file per question, slug-keyed
• answers/ — one markdown file per answer, same slug as the question
• context/ — long-lived framing, definitions, prior-art notes
• sources/ — logged sources referenced across answers
• outputs/ — consolidated exports, PDFs, reports
• notes/ — scratch thinking

Primitives (installed globally by the plugin)

• /research-space:research-init — load state at session start
• /research-space:source-log <ref> — log a new source
• /research-space:summarize-sources — per-source summaries + synthesis rollup
• /research-space:deep-dive <topic> — structured investigation
• /research-space:research-status — workspace state report
• /research-space:export-report — compose a shareable or consolidated report