Claude Code
Facts
- Openness
- Closed, studied from outside
- Licence
- proprietary
- Language
- Not stated
- Pinned version
- Official documentation (the code is closed)
- Topics checked
- 33 of 33
- Kind of agent
- Coding
What it does notably well
- Checkpoints and rewind
The rewind menu doubles as targeted compaction: Summarize from here or up to here compresses part of the conversation from a chosen checkpoint, leaving files untouched and the…
In the Agent SDK a checkpoint is the UUID of a user message in the response stream, so an application can keep one restore point or many and rewind to any of them after resuming…
- Compaction
After a summary, Claude Code rebuilds working state from disk: CLAUDE.md, auto memory and the plan are re-injected from disk and up to five recently modified files are re-read.
- Extension hooks
PreToolUse hooks can allow, deny, ask or defer a tool call and can rewrite its input, with deny, defer, ask, allow precedence when several hooks disagree, and the rewritten input…
- MCP and auth
Only MCP tool names and server instructions load at session start; schemas load on demand through ToolSearch, so adding servers has minimal effect on the context window.
- Prompt caching
Cache-aware UX: /model asks for confirmation only while the cache is warm (within one TTL) and skips the prompt once the cache has expired.
See how Claude Code does on each topic
The topic-by-topic verdicts, the code lines and a side-by-side comparison are in the app. Sign in with Google; new accounts are approved by hand.
Independent analysis, not affiliated with Claude Code.