
What happened
The developer behind zero-trust agent tool ACP ended its development, renamed the repo to sealbridge, and published sealbridge on npm as a CLI and MCP server for delegating tasks to another owner's agent.
Why it matters
The redesign shifts distribution from cloning a Git repository to a one-command install, and replaces manual key exchange with an invite link—removing the two biggest hurdles to trying it out, based on the developer's account.
Summaries like this, every morning.
Sealbridge is the successor to ACP, a zero-trust framework the same developer previously released as open source for delegating work to others' Claude Code or Codex without handing over API keys. The rewrite changed both the design and the distribution model: ACP required cloning a Git repository, while sealbridge is an npm package installed with a single command, which the developer says dramatically shortens the distance to getting someone else to try it. The way relationships start also changed—from cumbersome key and connection-info exchange, which the developer identifies as the biggest drop-off point, to simply passing an invite link or token. The approval model is deliberately conservative: agents automatically accept only read-only work from connected partners, while write or command-execution tasks need the owner's approval, and running a worker on one's own PC requires an explicit --allow-local-pc flag (fail-closed). Every request, acceptance, start, delivery, verification, and completion event is signed and chained, so tampering with one side's log can be detected by comparing it with the other; when no trust anchor is present, the tool honestly reports that truncation detection is not guaranteed. As a real-world test, the developer delegated readability reviews of text to an agent owned by someone else (Grok Bot). After several days of isolating unstable relay connections caused by a proxy environment, a delegated task reached SETTLED in about 12 seconds, the deliverable was returned, and the audit log verified as VALID with no retries—showing that asking another person's agent for work and receiving it with evidence can be a routine operation rather than a demo.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Google's Gemini 4 Carbon took the #1 spot on MathArena with an expected 85.4% score, displacing OpenAI's GPT-6…

In a test of PurpleDoubleD's locally-uncensored app (AGPL-3.0-only), a coding agent was told to fix a one-line…

On November 10, 2026, Anthropic released Claude Sheets in beta (building spreadsheets from prompts, 144 suppor…

A practitioner writes that after enriching specs and CLAUDE.md rules, AI still produced off-intent behavior, a…

The author fed the 'desired candidate profile' text from 145 job postings to Claude, turned the extracted labe…

Researchers de-duplicating leak-site data say alleged ransomware attacks reached nearly 3,000 over the summer…
