Skip to main content
← Back to list
01Issue
FeatureShippedExtensionsPublicTeam
Assigneesstack72

Relationships

↔ sibling #4764↔ sibling #4747

#4768 @swamp/ssh: a model that calls a fleet can't tell whether two fleets reach a host by the same route

Opened by randybias · 11/15/2026· Shipped 11/15/2026

Type: feature request.

Versions: @swamp/ssh 2026.10.23.1, swamp 20261114.210101.0.

What happens: two @swamp/ssh fleet definitions can reach the same host by different routes: through a different jump host or proxy command, from a different local bind address, or with ControlMaster on in one and off in the other. A model that drives a fleet through context.runModel cannot find out which route applies:

  • resolve returns each matched host's name, address, port, user, tags, attributes and transport kind, but nothing about the proxy, bind address or connection sharing;
  • the host records written by apply carry the same fields, again without them;
  • a model cannot read another definition's arguments, so it cannot look the settings up there either.

What we expected: resolve's hosts and the host records also report, for each ssh host, the effective route settings, as the connection will actually use them (fleet defaults merged with the host's own):

  • the effective proxy as a digest (a hash of the effective ProxyJump and ProxyCommand), never the values, because a proxy command can carry credentials; left out when no proxy is set;
  • the bind address (the first BindAddress option), left out when there is none;
  • whether ControlMaster is enabled;
  • the control options (ControlPath, ControlMaster, ControlPersist, -S) given as extra options, left out when there are none.

Tailscale hosts have none of these.

Shortest reproduction:

  1. Define two @swamp/ssh fleets with the same host, one with a proxyJump and one without.
  2. From another model, call each fleet's resolve through context.runModel and read the selection records.
  3. The two records are identical for the host: nothing tells the model the routes differ.

Related: Lab #4764 asks for the host-key policy in the same records.

Why it matters: a firewall change that must not lock the operator out proves reachability from a second connection before it commits. That proves nothing if the second connection takes another route, for example around the jump host the operator really uses, or from another source address. The model needs to see that both fleets reach the host the same way, and to refuse when they do not. Today it has to trust the caller, and a restated value proves nothing.

Upstream repository: https://github.com/swamp-club/swamp-extensions

Environment

  • Extension: @swamp/ssh@2026.10.23.1
  • swamp: 20261114.210101.0-sha.537e03d1
  • OS: linux (x86_64)
  • Deno: 2.9.8
  • Shell: /bin/bash
02Bog Flow
✓OPEN✓TRIAGED✓IN PROGRESS✓SHIPPED+ 3 MOREISSUE_LINKED+ 5 MOREREVIEW+ 10 MOREPR_MERGED+ 2 MORESESSION_SUMMARIZED

Shipped

11/15/2026, 10:39:35 PM

Click a lifecycle step above to view its details.

03Sludge Pulse
stack72 assigned stack7211/15/2026, 7:38:25 PM
stack72 linked sibling of #476411/15/2026, 7:38:50 PM
stack72 linked sibling of #474711/15/2026, 7:38:53 PM
Editable. Press Enter to edit.

system commented 11/15/2026, 10:29:33 AM

Classified automatically when this issue was filed.

  • Source: Extensions

If you feel this classification is incorrect, add a ripple to tell us so.

stack72 commented 11/15/2026, 10:39:39 PM

Thanks @randybias for reporting this! We shipped: Report each ssh host's effective connection facts in resolve's selection records and apply's host records: a SHA-256 digest of the proxy settings, the first BindAddress, whether ControlMaster is on, control extra options (#4768), and the effective host-key policy (#4764). Fix exec and script timeouts with ControlMaster on, which waited for the remote command because the master holds the output and input pipes (#4747).. The fix has been merged and a release is on its way. We appreciate your contribution to swamp.

Sign in to post a ripple.