Claude API Skill Onboards Managed Agents Fast
The fastest way to use the new API credits is through Claude Managed Agents. Running /claude-api managed-agents-onboard inside Claude Code walks users through setup.
ClaudeDevs·Thu, Oct 8 12:09pm ET
Monthly API Credits Land for Max and Team Plans
Max and Team subscribers now get monthly Claude Platform API credits: $100 for Max 5x, $200 for Max 20x, and up to $500 pooled for Team plans. Credits work across any Claude model, including Haiku 5.5, in your own code or third-party tooling.
ClaudeDevs·Thu, Oct 8 12:09pm ET
Sonnet 5.5 Cache Read Prices Cut in Half
Cache read pricing for Sonnet 5.5 on the Claude API drops to $0.10 per million tokens, down from $0.20. Combined with existing input ($2/M) and output ($10/M) rates, this makes most agentic workloads roughly 20% cheaper. The change is API-only and does not affect Claude Code usage limits.
ClaudeDevs·Thu, Oct 8 4:04pm ET
Claude Projects Waitlist Now Open
The new Claude Projects feature is rolling out gradually. A waitlist is available for those who want early access.
dfeinition·Wed, Oct 7 6:11pm ET
Claude Projects Promotes Thread Insights Project-Wide
Work starts in a main chat, branches into focused threads, and when a thread insight applies globally, Claude can elevate it to a project-wide instruction automatically.
dfeinition·Wed, Oct 7 6:11pm ET
Claude Projects Learn From Natural Conversation Flow
A workflow where project-wide instructions emerge organically: start with open-ended jam sessions in the main chat, move specific topics to threads, then promote anything broadly applicable into project-level instructions. The result is a system that feels less like prompt engineering and more like teaching through conversation.
dfeinition·Wed, Oct 7 6:11pm ET
Claude Projects Compounds Context Across Sessions
As project knowledge accumulates, shorter prompts produce better results because Claude already knows the context. The richer the project memory, the less you need to re-explain.
dfeinition·Wed, Oct 7 6:11pm ET
More Stories
Claude Projects Updates Docs When Plans Change
When working across many threads and plans shift, you can tell Claude what changed and it figures out which project documents are affected and updates them accordingly.
dfeinition·Wed, Oct 7 6:11pm ET
Claude Can Learn From Cross-Tool Team Comments
A useful trick for Claude Projects: prompt Claude to read your team's comments across external docs and tools, then internalize that context as project knowledge.
dfeinition·Wed, Oct 7 6:11pm ET
Claude Dashboards and Motion Launch in Beta
Two new Claude capabilities are now in beta: Dashboards turns your data into live, interactive dashboards, and Motion converts ideas into animated explainers. An Anthropic team member's reaction suggests internal enthusiasm.
noahzweben·Thu, Oct 8 4:19pm ET
Anthropic Debunks CIA Data Leak Claim
A user claimed Claude Code was leaking data to the US government after finding .gov strings in the Claude Desktop binary. Anthropic's Brian Cherny clarified these are standard AWS region endpoints from the AWS SDK used for Bedrock customers, not government surveillance infrastructure.
bcherny·Thu, Oct 8 4:05pm ET
Opus 5.5 Usage Draining 2-4x Faster After Update
A user on the Max 20x plan noticed their Opus 5.5 (high) usage depleting two to four times faster since a recent rollout. An Anthropic team member asked for a /feedback session ID or account email to investigate.
lydiahallie·Thu, Oct 8 2:18pm ET
Imprecise Prompting Is the Most Common Agent Failure
Responding to a discussion about programmer vs. non-programmer vibe-coding advantages, trq212 identifies the real bottleneck: people working outside their domain who can't be precise with prompts and plans end up burning many turns iterating vaguely.
trq212·Wed, Oct 7 10:26pm ET
Ask the Model to Teach You What You Don't Know
Following a point about imprecise prompting being the top failure mode, trq212 notes that because agents lower the barrier to entry, nearly everyone eventually works outside their domain. The practical fix: just ask the model to explain the unfamiliar concepts directly.
trq212·Wed, Oct 7 10:26pm ET