Run an on-demand session health audit — checks for spec drift, sycophantic confirmation, agentic loops, context degradation, silent failures, and premature completion
From agentic-watchdognpx claudepluginhub nbkm8y5/claude-plugins --plugin agentic-watchdog/health-checkImplements database health monitoring for PostgreSQL and MySQL with real-time metrics, predictive alerts, automated remediation, and Grafana dashboards using Prometheus exporters.
/health-checkCreates health check scripts to verify service and infrastructure availability, with standardized status format, alerting setup, scheduling via cron or Kubernetes probes, and documentation.
/health-checkRuns a comprehensive health check: auto-detects relevant auditors, runs them in parallel, produces a unified report. Optional exclusions (e.g., 'skip memory skip energy').
/health-checkAssesses .NET solution health across 6 dimensions—build, anti-patterns, architecture, tests, dead code, dependencies—producing letter grades (A-F), scores, and actionable report.
/health-checkChecks etcd cluster health, member status, leadership, connectivity, and identifies issues in OpenShift.
/health-checkAnalyzes and grades component health for an OpenShift release using regression metrics and JIRA bug data, with automatic data fetching and filtering by components or team.
Perform a comprehensive session health audit across all six failure dimensions.
Load skill and heuristics:
${CLAUDE_PLUGIN_ROOT}/skills/failure-detection/SKILL.md for the failure pattern overview.${CLAUDE_PLUGIN_ROOT}/skills/failure-detection/references/detection_heuristics.md for the specific signals to evaluate.${CLAUDE_PLUGIN_ROOT}/skills/failure-detection/references/severity_schema.md for severity levels and report format.Load previous report (if exists):
.claude/watchdog-report.md if it exists. Note any previously flagged issues to check if they are resolved or worsening.Evaluate all six dimensions:
Spec Drift: If artifacts/ directory or any spec file exists, compare recent code changes against spec requirements. Look for files with no spec coverage and requirements with no code. Assign healthy/warning/critical.
Sycophantic Confirmation: Review the current session for claims of success. For each claim, verify a corresponding Bash command ran the actual test/build/lint. Check for agreement with contradictory requirements. Assign severity.
Silent Failures: Review recent Bash command outputs for error indicators that may have been overlooked — errors in passing builds, zero-test runs, permission issues. Assign severity.
Tool Selection: Check recent tool usage for suboptimal patterns — Bash used for tasks that have dedicated tools, Write used where Edit would suffice. Assign severity.
Agentic Loops: Check for repeated edits to the same file (3+ times), repeated Bash commands, or alternating approaches. Assign severity.
Context Degradation: Test your own recall of the original requirements or task. Can you list all requirements? Have you contradicted any earlier decisions? Assign severity.
Write report:
.claude/ directory if it does not exist..claude/watchdog-report.md following the template in severity_schema.md./health-check.Present summary:
A summary of the health check results with per-dimension severity ratings, followed by specific recommendations for any warning or critical findings. The full report is persisted at .claude/watchdog-report.md.