skills.rest
curtismu7
curtismu7COMMUNITY
Explore all skills by this author →
GitHub

agent-demo-triage

Diagnoses silent agent pipeline failures using log signatures and root-cause tables.

1|Updated Jun 25, 2026
Search Tags:#troubleshooting#ai agent#log analysis#docker compose#feature flags#mcp gateway#pingone
One-click install
npx skills add https://github.com/curtismu7/AI-DEMO2 --skill agent-demo-triage-curtismu7
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agent-demo-triage
Source: https://github.com/curtismu7/AI-DEMO2/tree/main/.claude/skills/agent-demo-triage
Command: npx skills add https://github.com/curtismu7/AI-DEMO2 --skill agent-demo-triage-curtismu7

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? The banking demo agent pipeline fails silently: most breakages surface as a polite "Heuristics-only mode" catalog card instead of an error, making root causes nearly impossible to spot from the UI. This Skill maps each failure signature to its verified root cause and fix so you stop guessing. ## Core Features & Use Cases - Signature-to-fix lookup table: Matches log/UI symptoms (catalog card, 502/503 from /api/mcp/tool, "Unknown action" toasts, empty Code Explorer answers) to confirmed root causes like stale agent-service images, ctx-size overflow, feature-flag drift, and vault key mismatches. - Verification commands: Provides Playwright proof suites and an offline use-case trigger audit to confirm fixes before shipping. - Deployment invariants checklist: Lists containers needing rebuilds, flags that must hold (ff_heuristic_enabled, ff_authorize_real), and llama backend health checks. - Use Case: When a use-case chip phrase returns the catalog card in only one vertical, the Skill directs you to add perVertical config in useCases.js and re-run the offline trigger audit. ## Quick Start Ask the agent to diagnose why the demo replies with the heuristics-only catalog card instead of an LLM answer.

Frequently Asked Questions about agent-demo-triage

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
Why does the agent reply with the heuristics-only catalog card instead of an LLM answer?▼

The catalog card is a fallback for at least five distinct root causes, including a stale agent-service image, llama context-size overflow, or the ff_heuristic_enabled flag being off. Always check docker logs for agent-service, api-server, ping-gateway, and authz-server rather than trusting the rendered reply.

How do I fix 502 'Insufficient scope' errors on write tools like create_transfer?▼

The gateway edge exchange is requesting read-only scopes. Set PG_OLB_SCOPE to 'read write mcp:invoke' in ping-gateway/.env, restart ping-gateway, and re-assert the ff_gateway_brokered_exchange flag to avoid BFF/gateway split-brain.

What causes a 503 mcp_authorize_unavailable error from /api/mcp/tool?▼

The authorize backend is unreachable and failover is set to deny. The preferred fix is ensuring real PingOne Authorize is provisioned with ff_authorize_real=true; the simulated authz-server fallback is a last resort, not the default.

Why does a use-case chip phrase work in one vertical but fail in another?▼

The use case is missing a perVertical mapping, so a banking-specific phrase leaks into other verticals. Add perVertical config in config/useCases.js and re-run the offline trigger audit until it prints 'all triggers match'.

When should I rebuild containers instead of relying on hot reload?▼

Containers without src mounts — agent-service, authz-server, mcp-server, and ping-gateway — require docker compose build after any code change. Only the BFF and UI hot-reload from the checkout, and the BFF's node --watch causes a ~35s restart outage.

Related Skills

Explore More Software Engineering Skills→
topoteretestopoteretes
30.4k

cognee-docker

Deploy the cognee API server and full stack using Docker images and compose profiles.

Official
Intermediate
coopeverythingcoopeverything

verify-fix

Map code changes to Playwright E2E tests and execute them against production.

Community
Advanced
neuron-oneneuron-one
25

monitoring

Automate application and server monitoring with error tracking, metrics, and alerting.

Community
Intermediate
naviktnavikt
1

observability-setup

Configure Prometheus metrics, OpenTelemetry tracing, and health endpoints for NAIS applications.

Official
Advanced