The loop
- Locate the version. Read the installed package docs for every StackNext
package involved. Confirm the version in
package.json. - Confirm the sample. Call
get_project_summary; check connection, retention window, counts and coverage. - Choose one question. Pick an error, a slow route, a query pattern, a diagnostic finding, a comparison or a snapshot.
- Collect bounded evidence. Use list tools with a narrow filter, then fetch
one record by id. Paginate with
offset; do not dump everything. - Verify correlation. Exact span context is stronger than the route/time fallback. A verified regression sample requires the injected trace and a parented server span.
- State what is unproven. Diagnostics are heuristics; comparisons are not statistical tests; snapshots are not live.
- Recommend one next step. Prefer a measurement or a targeted request over a large refactor.
Error
list_errors → get_error → get_request / get_trace →
stacknext_list_diagnostics with category: "errors". Report the error id,
route, status and whether the trace link is exact or fallback.
Slow request
find_slow_requests → get_request → compare self time and query evidence.
Read slow-request and slow-span findings, but verify the contributor in the
trace. Do not call a null-duration query slow.
Diagnostics
stacknext_list_diagnostics → stacknext_get_diagnostic. Read counts,
sample, notes and limitations. Report the finding id, rule, status,
severity and the evidence actually checked.
Regression
stacknext_list_test_runs → stacknext_get_test_run →
stacknext_list_baselines / stacknext_get_baseline →
stacknext_list_comparisons / stacknext_get_comparison. Report status and
severity separately. Never create a comparison from telemetry.
Snapshot
stacknext_list_snapshots → stacknext_get_snapshot →
stacknext_get_snapshot_trace. State the snapshot id, source and save time.
Do not present it as current telemetry.