Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-03 · high
The same release also seals denied secrets out of the sandbox and bills for proven activity, not wall-clock time.
Why now: Suna shipped 0.12.0 on August 1 and a 0.12.1 patch the next day, per the project's release notes.
Suna 0.12 rewrote tool approval, secrets, and sandbox billing on August 1, per the project's release notes. That closes a real gap for teams that let agents run SQL or hit internal APIs. Approving run_sql once used to approve every future query, no matter what it ran.
Sessions start locked down by default. Connectors and secrets need explicit opt-in scope, agents can require specific connector profiles, and each profile carries its own authorization strategy. Secrets that get denied never enter the sandbox at all, so there's nothing left to leak later.
Approval prompts now show the full tool arguments, so whoever clicks approve can see exactly what's about to run. Suna also added a centralized audit log covering every API request and agent action with client attribution, plus per-session cost records.
Sandbox billing switched from wall-clock time to liveness evidence, charging for proof the sandbox was actually running rather than just elapsed minutes.
Each link below shares sources, entities, or timing with this story.
AWS announced and open-sourced Kiro Crew on August 4, Apache-2.0. It's a persistent multi-agent development workspace that coordinates coding agents across repositories, tools, and sessions rather than inside a single chat. Persistent memory. Scheduling via cron and webhooks....
3,364 stars since its August 17 creation. Every action against a computer, file, MCP server or UI component routes through a single gateway that resolves the target, decides it against policy, writes an audit row, then acts or refuses while naming the rule. Each bot gets its o...
v1.38.0 and v2.6.3, both August 7, port "prevent continuation forgery in tool confirmation," now validating that a confirmed target tool is registered in the executing agent's tools_dict, that it genuinely requires confirmation, and that the original tool-call event exists in...
PR #9379, merged August 6, fixes MCP tool entrypoints built as functools.partial(call_tool, tool_name=tool.name). Partial keyword defaults get overridden by call-time kwargs, so a model passing tool_name="delete_repo" as an argument to any MCP tool executed delete_repo while a...
The Rust sandbox runtime for autonomous agents is at 8,465 stars and cut v0.0.116 on August 28, still on a 0.0.x line five months after its February 24 creation (GitHub). A 1:2.8 PR-to-issue ratio reverses the PR-heavy pattern across every other trending agent repo this week....
rust-v0.150.1, published August 27, fixes remote compaction ignoring retained images when computing token budget; it now trims older images as needed (GitHub). Sessions passing screenshots to a vision-capable agent were budgeting as though those images cost nothing, which prod...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.