Fetching from the wire…
Public story · 2026-08-17 · high
1.59.0 rebuilds cost control with per-request spend limits and parks rule-proposed runs as approvals, not auto-execution.
Why now: Mastra shipped @mastra/core 1.59.0 on August 16, covered in yesterday's briefing.
Mastra's 1.59.0 release turns automatic agent runs off by default, requiring approval before a rule-proposed run executes. The change affects anyone running Mastra agents in production: workflows that used to fire automatically now wait on a person to approve them.
The mechanism is a new autoRunEnabled setting in Factory. When a rule proposes a run, that run gets parked as a "proposed" decision instead of firing immediately, per the GitHub release. Upgrading to 1.59.0 silently stops any workflow that depended on rules triggering runs on their own. Someone has to flip the flag back on or start approving proposed runs by hand.
The second breaking change replaces CostGuardProcessor with TokenCostControl. The new system adds a warnAtPercent threshold and per-request maxCost functions. It also splits spend limits into separate scopes for users, organizations, and sessions. That's a finer-grained way to catch spend before it runs away. It suggests the old blanket cost guard wasn't enough for teams running agents at scale.
A smaller change is worth flagging too. Trace redaction now uses indexed labels like [APIKEY_1] instead of one blanket [REDACTED] stamp. A trace with multiple secrets no longer hides which one is which.
The GitHub release doesn't say what specifically pushed the auto-run default to off, only that it happened.
Each link below shares sources, entities, or timing with this story.
The TypeScript framework added a v1 ToolProvider runtime for managing OAuth-backed connections, @mastra/mysql as a first-class storage backend, and a VercelMicroVMSandbox built on ephemeral Firecracker microVMs (GitHub). The MicroVM sandbox is the standout. Per-session isolate...
Frontier labs publish demos. This one published the thing they actually page. Anthropic's August 18 writeup describes Claude Tag running as the first responder for CI failures inside the company. Dedicated service account. MCP connectors to Datadog, Grafana, PagerDuty, GitHub...
The official marketplace now lists 200+ plugins, ~20 first-party, the rest vetted partners for GitHub, Supabase, Figma, Vercel. Discovery and curation, not authoring, is now the bottleneck. Audit which plugins you actually trust before installing broadly, because a plugin bund...
1. Use claude agents --json to build session dashboards. Claude Code v2.1.145 outputs all live agent sessions as structured JSON with status, model, elapsed time, and parent relationships. Pipe it into a tmux status bar widget or session picker script for switching between bac...
MCP server exposing 150+ cybersecurity tools for AI agents: nmap, sqlmap, ffuf, and 147 others. Covers pentesting, vulnerability discovery, and bug bounty automation. Works with Claude, GPT, Copilot — any MCP-compatible agent. Docker support included. Add it to your MCP config...
Also: stored agents from the Agents API can run with durable: true with no redeploy, @mastra/rag gained serialize()/GraphRAG.deserialize() so knowledge graphs survive restarts, and sandboxes advertise supportsCheckpoints with filesystem-backed snapshots. (GitHub) A day earlier...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.