Try
Useful tools & techniquesCustomize Claude Code with mods and workflow tools
Claude Code mods are available in the CLI and desktop app as TypeScript functions packaged in plugins. They can rewrite prompts, block or retry tool calls, change permissions, redact tool output, and alter the UI; mods have the same machine access as Claude Code, so install only trusted ones. The built-in /diff is now a mod, and the “You should Know” plugin scans output for overlooked details. One shared workflow proposes writing a short HANDOFF.md, clearing a long session, then switching models; its author estimates this can reduce recaching costs.
Try a handoff before switching models; review mod permissions and trust as carefully as installed code.
Claude edits Google Workspace files in beta
Claude for Google Workspace is in public beta on all paid Claude plans. Its Docs, Sheets, and Slides sidebar can read open files and edit them in place, with approval cards by default or an option to accept all edits. Beta connectors also let Claude create and edit Google files from chat. Team and Enterprise admins must enable connectors; Enterprise controls including CMEK and audit export apply to the add-on.
Use approval mode for edits; admins can deploy the add-on to selected groups.
Claude can join Slack group DMs
Claude can now be added to Slack group direct messages. It replies in a thread and continues following it; members can reply there, and Claude can use the personal connectors of the person asking.
Try it for shared questions, but remember connector access depends on who asks.
Know
What changedHaiku 5.5 cuts costs; Sonnet cache reads get cheaper
Claude Haiku 5.5 is available across Anthropic’s platform and major clouds for high-volume work, summaries, and coding subagents. Anthropic says it costs 90% less than Haiku 4.5 on prompts up to 100,000 tokens and 50% less above that. Sonnet 5.5 cache reads are now $0.10 per million tokens, half the former price. Max and Team subscribers also get monthly API credits; Python and TypeScript SDK support for computer and browser use is in beta. P-Zero Research claims Opus 5.5 matches expert research-task scores using 17 rather than 40 GPU hours.
Try Haiku for bounded, repetitive subtasks; keep complex agentic coding on Sonnet or Opus.
OpenAI begins text watermarking in the EU
OpenAI plans to add invisible text watermarks to eligible ChatGPT and Codex output in the EU over the coming weeks. API customers worldwide can opt in for select models; watermarking remains off by default. Detector access starts with approved researchers and expert organizations. OpenAI reports detection weakens with short, constrained, or edited text, so a result does not establish authorship, accuracy, or responsibility.
If you offer generated text, review the EU rollout and avoid treating watermark detection as proof of authorship.
Google announces open multimodal embedding model
Google DeepMind introduced EmbeddingGemma 2, an open model for on-device embeddings that represents code, images, audio, and video in a shared space. The announcement provides no further specifications or availability details in the captured post.
Watch for model details before choosing it for an on-device search or retrieval stack.
User alleges Opus 5.5 deleted a drive
A user posted that Opus 5.5 deleted their C: drive and said daily NAS backups saved their data. The captured post does not establish how the incident happened, but it is a reminder that computer-use agents can affect local files.
Keep backups current and limit an agent’s filesystem access, especially for unattended tasks.
Study reports AI outperforming junior accountants
Andrew Curran highlighted Mercor’s reported finding that frontier models were faster and more accurate than junior accountants on medium-length, well-defined accounting tasks. The post says models had scored below the human average 18 months earlier; the captured material does not provide the study’s methods or results tables.
Treat this as a task-specific benchmark claim, not evidence of general workplace performance.