A backpressured stdout write in the duplex gateway waited for drain or a stream error, with no other end condition. A pipe that stays open but stops draining -- for example a reader that stops consuming without closing the pipe -- never fires either event, so the wait, the send, and the admission reserve for that read stayed open forever. Later requests then answered 503 bridge_busy for good. Add a stall-timeout end condition to the embedded stdout frame writer. It reuses the existing response timeout as its bound: a write that stays backpressured that long already means no response can arrive in time, so ending the wait costs nothing beyond what the response side already gives up on. Once the bound elapses, the writer treats the wait exactly like a stream error -- it strands every queued line and refuses further writes -- so the same finally block that already releases the reserve on a stream error now also runs. Add a unit test that proves a stalled write past its bound rejects and strands the queue, and a companion test that proves a write which drains before the bound clears its timer and settles normally. Add an end-to-end test that pauses the real child's stdout pipe and never resumes or destroys it, proving a later request is admitted again once the stall bound elapses. Confirmed each new test fails against the prior commit and passes after this change. Co-authored-by: Paperclip <noreply@paperclip.ing>
@paperclipai/adapter-utils
Shared utilities for Paperclip adapters: process spawning, environment injection, sandbox/SSH transport, workspace sync, and the round-trip helpers that move code between the local execution-workspace cwd and wherever the agent actually runs.
For the adapter-author guide see
docs/adapters/creating-an-adapter.md
and the in-repo notes at packages/adapters/AUTHORING.md.
No-remote-git contract
The local execution-workspace cwd is the only persistence boundary across runs. No adapter may depend on a git remote for cross-run state.
Adapters that run the agent on a different host should use the SSH round-trip
helpers in src/ssh.ts:
prepareWorkspaceForSshExecution({ spec, localDir, remoteDir })— bundles the local cwd (tracked files, dirty edits, untracked additions, and the git history needed to reconstruct it) toremoteDirbefore the run starts. Runs with nogit remoteconfigured.restoreWorkspaceFromSshExecution({ spec, localDir, remoteDir, ... })— syncs the remote cwd back intolocalDirafter the run, including any new commits the agent created. Also runs with nogit remoteconfigured.
prepareRemoteManagedRuntime in
src/remote-managed-runtime.ts wraps both
calls for adapters that want a per-run remote workspace and an automatic
restoreWorkspace() finally hook.
The invariant is pinned by the no-remote-git contract case in
src/ssh-fixture.test.ts, which asserts that a
remote-only commit propagates to the local worktree through the
prepare → restore round-trip with no git remote configured at any point. Do
not regress that test.