be post ssh://…/<sub> forges a funny ref refs/heads/.beagle/
A mounted beagle submodule's current branch is the synthetic dot-coordinate ?/<sub>/.<parent>[/<br>] (be-only; see Submodules). Posting that worktree DIRECTLY to a git remote derives the dot-branch as the wire refname, so keeper sends refs/heads/.beagle/ and the peer rejects it ("refusing to update funny ref") — after a full pack build, ending in WIRECLFL/BEDOGEXIT. Recursion already strips the transport for subs (beagle/BE.cli.c:2086, commit 8e2a5b2a); the direct top-level post of a sub is unhandled. The goal: a dot-branch never reaches a git wire. See POST, Submodules.
The push refname is taken from cur_branch with no check that a dot-coordinate is be-only.
keeper/KEEP.exe.c:818-825 derives local_branch from h->cur_branch (= .beagle/) when the URI has no ?ref.wcli_be_to_wire (keeper/WIRECLI.c:314) → GITFeedRef wraps it blindly: refs/heads/.beagle/, no refname validation.cd into a mounted beagle sub (cur = dot-branch), be post ssh://localhost/src/dogs/<sub> → funny-ref reject.None.
Skip the wire push when the branch is a synthetic dot-coordinate and the remote is a git transport — local commit (sniff) already happened, nothing to push to a git peer.
DOGIsGitTransport(u8cs) to dog/DOG.c/DOG.h (ssh/https/http/git) beside DOGIsTransport.keeper_post (after branch is computed, before remote-URI/pack build): if branch[0][0]=='.' AND the resolved remote scheme is a git transport, print a one-line note and return OK — build NO pack.be:// dot-branch pushes working (beagle remote is not a git transport) and normal-branch git pushes unchanged.test/post/: sub on a dot-branch + be post ssh://…/<sub> exits clean, no funny-ref, no pack.