Welcome to the Agentic Age. No question is bad. No topic is off limits. We are here for you — to help every engineer ship faster, think bigger, and build with confidence.
Auto mode now GA on Bedrock — no opt-in flag
Opus 4.8 is the new Bedrock default
Subagents now run in the background by default
/doctor is now a full setup checkup (/checkup alias)
/cd adds directory path suggestions
MCP request_timeout_ms & OAuth refresh fixes
Memory leaks fixed: MCP stderr, LSP, edit cache
Session transcripts pruned up to 79x smaller
Bedrock streaming & AWS SSO auth fixes
Driving deterministic behavior with AI — same discipline, every engineer, every session.
How Claude acts. Standards and skills defined in .md files at different levels.
A framework for writing instructions. Claude takes instructions seriously when written a certain way.
Are my instructions working? See when Claude isn't following so we can tweak the format.
Cascading, layered instructions so Claude knows how to act — standards defined in .md files at every level.
Concise · Staff-Eng. Approved · Factual · Safe
Architecture · Known Issues · Complete
Shared Knowledge · Ent. Patterns · Skills · Schema · Domain Docs
~/.claude/CLAUDE.md # universal └── project/CLAUDE.md # project └── @includes & Skills └── memory docs └── referenced docs # domain
Keyword semantics — Claude pays attention when instructions are unambiguous.
5 enforcement mechanisms — how you define the instruction set matters.
MUST pass before [trigger]
Any unchecked → STOP
[condition] → [action]
- [ ] checklist item
Do not proceed / ask user
See your instructions drive AI decisions and tool usage in real time — so you can spot drift and tweak the format.
Visibly see how an instruction was used on each turn.
Shows when tools are used — e.g. an MCP server or Skill fired.
Side-by-side recording — same prompt, same model, one with the framework and one without.
Opus 4.8 on both — same model, same prompt. The only variable is the framework.
Same AI discipline. Every engineer. Every session.