Context
hex-ssh-mcp now has a materially stronger transfer engine, but correctness is still proven mostly by local smoke tests and mocked SFTP behavior.
Before expanding feature breadth further, we need evidence that the current upload/download path behaves correctly across real servers and across both OpenSSH extension and fallback paths.
This is the first follow-up after #36.
Why first
ssh2 durability extensions are optional, not universal
- current confidence is high locally, but cross-server evidence is still thin
- directory transfer, audit interpretation, and backend strategy are all safer after real interop coverage exists
What to do
- Build an integration matrix for
ssh-upload and ssh-download
- Cover at minimum:
- OpenSSH SFTP with OpenSSH durability extensions available
- a second SFTP backend with different behavior and no OpenSSH durability extensions
- Validate:
- overwrite reject / replace paths
- recursive remote mkdir
- timeout behavior
- verification failures
- durable finalize with extension fallback
- local allowlist and remote allowlist behavior
- Add repeatable CI-friendly fixtures or containers where feasible
- Document observed backend differences and supported guarantees
Schema/API changes
- No public tool schema changes required
- May need test harness/config additions for integration targets
Definition of Done
Related issues
- Built on:
#36
- Unblocks / informs:
#37, #39, #40, #41
Priority
Highest follow-up priority. This is the key evidence gap still missing.
Context
hex-ssh-mcpnow has a materially stronger transfer engine, but correctness is still proven mostly by local smoke tests and mocked SFTP behavior.Before expanding feature breadth further, we need evidence that the current upload/download path behaves correctly across real servers and across both OpenSSH extension and fallback paths.
This is the first follow-up after
#36.Why first
ssh2durability extensions are optional, not universalWhat to do
ssh-uploadandssh-downloadSchema/API changes
Definition of Done
Related issues
#36#37,#39,#40,#41Priority
Highest follow-up priority. This is the key evidence gap still missing.