Terminal
Terminal/Skill
Narrows Konnect Gateway failures such as disconnected data planes or missing rollouts to attachment, network, drift, or traffic.
By Kong
Works with
npx skills add kong/ai-marketplace --skill konnect-gateway-triageReduce ambiguous "gateway is broken" reports into a concrete failing layer: control plane state, data plane registration, network path, config drift, or traffic behavior.
Default to live inspection first. Use repo config only as intended state, not as proof of deployed behavior.
kong-konnect MCP server first for current control plane and
data plane state.kongctl-query for concise, read-only checks when CLI output is easier
to summarize or filter.deck-gateway for Gateway-entity
decK repos, terraform-konnect for HCL-managed Konnect Gateway
resources, terraform-kong-gateway for self-managed Gateway HCL, and
kongctl-declarative for kongctl YAML.kong-konnect MCP is not connected, say
so early and continue with CLI or user-provided artifacts.Prefer to inspect before suggesting restarts, reprovisioning, or config rewrites. Do not let this skill absorb generic upstream application debugging or the declarative mutation workflow itself.
Load only the reference file that matches the active branch:
references/connection-bootstrap.md
references/live-vs-declarative-drift.md
references/traffic-path-vs-control-plane.md
references/common-failure-signals.md
Classify the problem up front:
Do not investigate all paths at once. Pick the failing operator symptom and follow it.
If the symptom is still vague after this first pass, load
references/common-failure-signals.md.
Verify:
Many "gateway" issues are actually wrong-control-plane issues.
For disconnected or stale behavior, inspect data plane registration and current association first.
Look for:
If the data plane is not healthy, stop treating the problem as a config bug.
Load references/connection-bootstrap.md when registration or attachment is
the active failure branch.
If a data plane will not connect, isolate whether the blocker is:
Do not jump to service or route debugging until the control-plane connection is healthy.
Use references/connection-bootstrap.md for concrete bootstrap-versus-network
separation.
Once the control plane and data plane are healthy, inspect the live resources that should be deployed:
Prefer the smallest live slice that proves or disproves the suspected drift. If repo config disagrees with live state, call that drift explicitly. If live state is correct but traffic still fails, move to request-path troubleshooting instead of deployment troubleshooting.
Load references/live-vs-declarative-drift.md for repo-versus-live ownership
questions, or references/traffic-path-vs-control-plane.md when healthy
attachment still does not produce the expected traffic behavior.
Use one primary diagnosis:
This keeps the answer operational instead of producing a mixed list of guesses.
4xx or 5xx often means the
control plane is no longer the primary failure domain.Before answering, verify that you can state:
kongctl-query for exact resource inspection commands and filtered reads.deck-gateway, terraform-konnect, terraform-kong-gateway, or
kongctl-declarative when the operator wants to fix drift or apply planned
gateway changes. Match the repository's current toolchain.konnect-access-scope if the root problem becomes access to the control
plane rather than gateway health.konnect-observability-triage if the control plane is healthy and the
main question is missing analytics or debugging data.