The ant CLI brings every Claude Platform API endpoint to the terminal, covering the Messages API, Managed Agents, and shell piping. It's also natively understood by Claude Code via the claude-api skill.
Running "ant auth login" triggers a browser-based OAuth flow that scopes a token to a workspace and stores it locally. That single credential is then reused automatically by both the CLI and the Anthropic SDKs.
Every Claude API resource, including messages, models, and files, is a typed subcommand in the ant CLI. Requests can be built from flags or piped YAML, files inlined with @.path syntax, and output rendered as JSON, YAML, JSONL, or an interactive fold-and-search TUI.
The ant CLI can start a Managed Agents session, send it events, and later pull a complete trace of every tool call, reasoning step, and decision once the run finishes or stalls.
Agents can be defined as YAML, committed to Git, and synced to the Claude Platform automatically with "ant beta:agents update" in CI, making agent configuration a first-class part of a normal software workflow.
Workflows are framed as the most significant capability expansion to Claude Code since skills and subagents shipped. A deep dive with best practices and examples covers the feature, with particular emphasis on non-technical use cases it unlocks.
You can encode your usual manual review steps directly into Claude's workflow, letting it close the feedback loop itself instead of handing back unverified output.
A bare link posted by an Anthropic team member, likely pointing to the workflows guide or blog post. No additional context provided in the tweet itself.
The workflows deep dive is cross-posted to the official Claude Blog, making it available beyond wherever it was originally published.
When Claude unexpectedly switches to Haiku mid-task, you can prompt it to always inherit the parent model setting, which keeps it on Opus 4.8 instead of downgrading on its own.
When no model is explicitly set, the system inherits from the parent context. In this case, Claude itself chose to use Haiku.
Sonnet handles the majority of steps in a typical agentic workflow just fine. Opus isn't always necessary, and the distinction probably deserved more emphasis.
Structuring Claude workflows maps closely to multithreaded programming: a serial step queues independent tasks to run in parallel, with minimal cross-dependencies, followed by a serial post-processing step on the results.
Workflows are being positioned as a significant capability, with a linked article explaining what they are and why they matter.
Subagents start fresh with only their initial prompt, system prompt, and tools. A fork is a special kind of subagent that inherits the full conversation history, system prompt, tools, and model, making it essentially a clone of the parent session.
/fork runs a background agent carrying your exact system prompt, tools, history, model, and prompt cache, then returns the result to your session. The old /fork behavior (copying the transcript to a new session you control) is now called /branch.
The JS environment gets some determinism treatment, but it's not full temporal durability. A reasonable mental model: each subagent is roughly as durable as hitting /resume, not a persistent long-running process.
The keyword that activates ultra mode has been updated to 'ultra code.' A team member confirmed the change directly in reply to a user question.
