@dai_ming/plugin-agent-diagnosis
v0.1.5
Published
OpenClaw private agent diagnosis plugin — let an owner agent diagnose OpenClaw pods and no-reply issues through gateway self-diagnosis APIs
Readme
Agent Diagnosis Plugin
Private OpenClaw native plugin for an owner's main agent. It exposes diagnosis tools that call the gateway self-diagnosis APIs, so the agent can explain why a pod is unhealthy or why a message did not get a reply.
This plugin is intentionally not added to the default plugin set. Install it only for the owner's main agent or owner-specific managed image.
Runtime Config
Set one of these authentication modes in the target OpenClaw pod:
AGENT_DIAGNOSIS_MANAGE_TOKEN: existing manage JWT token.AGENT_DIAGNOSIS_MANAGE_USERNAME+AGENT_DIAGNOSIS_MANAGE_PASSWORD: login credentials used to fetch a short-lived manage JWT.
Optional:
AGENT_DIAGNOSIS_GATEWAY_URL: gateway base URL. Defaults tohttp://claw-gateway:8080.AGENT_DIAGNOSIS_DEFAULT_RELEASE: release name used when a tool call omitsrelease_name.AGENT_DIAGNOSIS_NAMESPACE: optional namespace used when resolving a display name through gateway pod mappings.AGENT_DIAGNOSIS_TIMEOUT_MS: HTTP timeout. Defaults to120000.
Tools
agent-diagnosis__statusagent-diagnosis__diagnose_unreplied_messageagent-diagnosis__diagnose_release_logs
release_name may be either the exact Helm release or a pod mapping display
name such as an owner nickname, agent nickname, or lobster IM ID. Non-release
names are resolved via
/openclaw-gateway/manage/api/pod-mappings before diagnosis.
By default tool results do not include raw logs. Pass include_logs: true only
when the agent needs short sanitized log tails for deeper troubleshooting.
