xterm and keydown can both emit carriage return for one keypress; skip duplicate Enter frames on the client and collapse identical audits within 450ms on the server.
Co-authored-by: Cursor <cursoragent@cursor.com>
Do not reuse the previous command when the operator presses Enter on a blank prompt; validate prompt-only rows on both frontend and backend.
Co-authored-by: Cursor <cursoragent@cursor.com>
PTY stdout/stdin carry backspace and edit noise during typing; record only the final row the operator sees when pressing Enter.
Co-authored-by: Cursor <cursoragent@cursor.com>
ZTE tab redraws the input line via carriage return; parse the last prompt line from PTY stdout on Enter instead of stdin keystrokes. Also remove remaining ?? in AuditPage.
Co-authored-by: Cursor <cursoragent@cursor.com>
Remove duplicate serializeTerminal fragment that broke WebTerminal compilation; rewrite ne.exec_batch count without mixing ?? and ||.
Co-authored-by: Cursor <cursoragent@cursor.com>
On Enter, send audit_line from the terminal row so tab completion and edits match what the operator executed; keep the device prompt prefix in audit logs.
Co-authored-by: Cursor <cursoragent@cursor.com>
Keep one 401 audit per IP+path window, and hide historical unauthorized noise from the default business view.
Co-authored-by: Cursor <cursoragent@cursor.com>
Persist only business events and mutating/failed HTTP calls, and default the audit UI to a business view with an explicit HTTP filter.
Co-authored-by: Cursor <cursoragent@cursor.com>
Make session lifecycle and typed CLI lines searchable in audit_log with password redaction, and surface summaries plus device filters in the ops UI.
Co-authored-by: Cursor <cursoragent@cursor.com>
Show connect/alarm/level on the canvas with display toggles, refresh overlay fields even on dirty edits, fix connect-test dots and create-place messaging, and deep-link node alarms into the network alarms page.
Co-authored-by: Cursor <cursoragent@cursor.com>
Bring full managed-NE create onto the canvas (keep placeholders), add single/multi connect test and multi discover links, split topology/managedNe dialogs into modules, and shorten redundant connectivity failure details.
Co-authored-by: Cursor <cursoragent@cursor.com>
ZTE post-login banners can line-wrap login-failure stats and were misclassified before the target prompt was recognized.
Co-authored-by: Cursor <cursoragent@cursor.com>
LLDP discover and config collect were timing out on --More--; send vendor paging-off after nested/bastion auth, and attach session-log tails so failures show where the CLI stuck.
Co-authored-by: Cursor <cursoragent@cursor.com>
Huawei NEs still tagged zte_zxros were getting show lldp neighbor brief and timing out; use display lldp neighbor and align UME inventory vendor before the jump.
Co-authored-by: Cursor <cursoragent@cursor.com>
Some hops only accept stelnet in system-view; make it a saved yes/no setting (default off) instead of auto-retry, and always ALTER the new column on API boot when Alembic skips legacy DDL.
Co-authored-by: Cursor <cursoragent@cursor.com>
Skip Huawei special_login_handler under interactive WebCRT, harden stelnet host-key auth, and echo hop/login progress into the terminal during connect.
Co-authored-by: Cursor <cursoragent@cursor.com>
Use the stacked form-grid layout instead of the nowrap policy-field flex row so the long label/hint no longer pushes the textarea aside.
Co-authored-by: Cursor <cursoragent@cursor.com>
Topology MCP adds pull/compact/bundle/suggest-hubs with a no-template skill path; API gains fabric level and NE collection policy; web list pages get paging and denser collect/network workflows.
Co-authored-by: Cursor <cursoragent@cursor.com>
Reject non-interactive CLI work early across exec, LLDP, config sync, collection, traffic, and connect-test flows so tasks record a clear reason instead of launching a doomed session.
Co-authored-by: Cursor <cursoragent@cursor.com>
Mirror config-sync retry_failed so operators can re-collect hard failures and empty/stub LLDP results without re-running the full scope.
Co-authored-by: Cursor <cursoragent@cursor.com>
Ensure startup safety-net adds a_ifname/z_ifname on brownfield DBs, and normalize ADD COLUMN IF NOT EXISTS for SQLite tests.
Co-authored-by: Cursor <cursoragent@cursor.com>
execManagedNe can take ne_ids/ume_ne_ids/targets and POST /v1/managed-ne/exec-batch so many show commands run in parallel instead of serial one-NE MCP loops.
Co-authored-by: Cursor <cursoragent@cursor.com>
Agents often passed UME UUIDs into getManagedNe and hit SQL without scope; aliases, richer hints, and scope error text reduce these dead ends.
Co-authored-by: Cursor <cursoragent@cursor.com>
Align MCP CLI calls with the API default so slow show commands are less likely to fail before the device responds.
Co-authored-by: Cursor <cursoragent@cursor.com>
Agents frequently passed group_by to the fixed aggregate tool; route those calls to raw aggregation and discourage list/exec retry loops.
Co-authored-by: Cursor <cursoragent@cursor.com>
Only expand adjacency for visited nodes, and rank objective=total with crossing 0.18 + clearance 0.08 (fallback to crossings when clearance is skipped).
Co-authored-by: Cursor <cursoragent@cursor.com>