| name | debug-integration |
| description | Figure out why a Resolve integration is failing or quiet, or explain what its settings do. Use when the user mentions a Resolve integration failing or wants to understand its configuration — phrases like "debug my <X> integration", "what's wrong with my Datadog", "why is my Tempo quiet", "my satellite isn't sending data", "explain the fields on my Grafana integration", or "what does <field> do on my <integration>". Read-only. |
| version | 0.1.0 |
| argument-hint | <integration-name or question> |
| license | Apache-2.0 |
Debug a Resolve integration
Use when the user wants Resolve to diagnose or explain one of their integrations. The Resolve agent has internal tools to list integrations, load schema + redacted values + last health check + satellite scope, and (for satellite-backed integrations) read pod logs scoped to the right namespace.
This skill is read-only. To create or mutate integrations, see resolve-ai-admin:create-integration. To edit satellite Helm values, see resolve-ai-admin:satellite-configs.
Arguments
If $ARGUMENTS is non-empty, treat it as the user's question and proceed.
If $ARGUMENTS is empty (skill invoked without a clear question), ask the user once what's wrong with which integration. Treat their next reply as the base message.
Composing the ask
Compose one message for resolve-ai:ask. The agent will discover the target integration internally (list_integrations → get_integration_details) and invoke its own integration-debug skill to apply the diagnostic discipline (kubectl scoping rules, structured output formats).
You don't need to specify the integration instance id — describing the integration by name and symptom is sufficient. The agent will resolve which instance the user means.
Enrich the base message with conversation context that directly supports debugging this integration. Be ruthlessly selective:
- Error messages, log lines, or stack traces the user has shown that name the symptom
- Recent changes (deploys, config edits) that might be tied to the failure
- Health-check output the user pasted, if any
Format the enriched message as the user's question first, then a brief ## Local context block with only the relevant items. Hand that single composed message to resolve-ai:ask; it owns sending, scoping, streaming, and canvas URL handling.
After asking
Once resolve-ai:ask has produced its streamed or settled answer, expect a Mode A 4-block diagnosis (starts with **What's failing:** and includes Evidence / Likely cause / Suggested next step) when the user asked about a failure, or a Mode B field tour when they asked about config fields. Surface this verbatim — don't paraphrase the structured blocks.
Out of scope for this skill
- Creating an integration →
resolve-ai-admin:create-integration.
- Editing an integration's connection config → currently the REST API (see
resolve-ai-admin:create-integration for the link) or the Resolve UI. Agent-mediated updates are not yet supported.
- Editing satellite Helm values →
resolve-ai-admin:satellite-configs.
- Investigating production incidents that aren't specifically about an integration → Resolve's investigate workflow (the
resolve-ai plugin, if installed).
If the user's question doesn't fit integration debugging, redirect to the right skill and stop.