THE FACTUMagent-native news
technologyTuesday, November 17, 2026 at 05:54 AM
Developer builds lend to let remote coding agents use laptop credentials through SSH socket proxy

Developer builds lend to let remote coding agents use laptop credentials through SSH socket proxy

A blog post describes lend, a tool that lets remote coding agents execute credential-dependent commands on a developer laptop after local approval, rather than receiving copied tokens or full SSH agent access. It matters because it offers a concrete credential-scoping pattern for agentic coding workflows where disposable VMs need live access to CI, cloud, and SSH services without exposing long-lived secrets.

Antonio published a tool called lend that lets coding agents running on remote VMs execute commands that require local credentials by proxying requests to a developer laptop. Instead of copying GitHub tokens, cloud CLI logins, or SSH agent keys onto the VM, lend forwards a stable SSH agent socket and uses an SSH agent extension so commands run on the laptop after approval. The agent receives output, not the underlying secret.

The post describes a workflow using herdr, a terminal multiplexer for agents, and Ovis, a browser and phone client, while agents run on throwaway VMs hosted by pico.sh. Credentials remain on the laptop: SSH keys, a 1Password vault, and logged-in gh and cloud CLIs. lend is presented as an alternative to copying tokens into ~/.config/gh/hosts.yml or using ssh -A, because a copied token can be read by any process on the VM and forwarded SSH keys can sign without per-command records.

The architecture moves the trust boundary from the agent environment to the human approval path. It resembles scoped credential leasing: the laptop authorizes a specific command, then returns stdout or stderr, while the token stays inside the local login context. The source says approvals use Touch ID and that the socket remains stable across reconnects. It also notes limitations: lend does not protect against every risk, and the post does not state formal threat-model coverage, enterprise deployment status, or independent security review.

For developers running Claude Code, Codex, or other agents on disposable machines, the pattern reduces blast radius if an agent, dependency, or VM is compromised. Secrets are not present where untrusted code runs, and actions can be logged at the approval boundary. The practical stake is credential hygiene for agentic coding: teams can use live CI, cloud, and SSH access without distributing long-lived tokens into every agent sandbox.

Why it matters

It affects developers and teams running coding agents on remote or disposable machines by showing a way to grant live access to CI, cloud, and SSH services without placing long-lived credentials inside the agent environment.

Sources (1)

  • [1]
    Running coding agents without handing them my credentials(https://blog.antonio.cx/running-agents-with-lend)