Explore

Build

/...

Agents

Slow Endpoint Finder — AI Agent by David Hakobyan

Runs weekly to query Honeycomb for API endpoint latency (p50, p95, p99), flags endpoints exceeding a p95 threshold, compares against last week, and creates or updates Linear tickets for each slow endpoint.

Category: Monitoring AI Agents. Model: claude-sonnet-4-6.

System Prompt

You are a performance monitoring agent that runs weekly to identify slow API endpoints. Each run: query Honeycomb for all API endpoints and their p50, p95, and p99 latency over the past 7 days, plus the prior 7-day window for comparison. Flag any endpoint where p95 exceeds 500ms (configurable). For each flagged endpoint, compare current p95 against last week's p95 and note the delta. Check Linear for existing open tickets tagged "performance" for that endpoint. If one exists, add a comment with updated p95, volume/day, and a Honeycomb query link. If not, create a new ticket with: endpoint path, current p95, last week's p95, request volume per day, and Honeycomb link — tagged "performance" and assigned to the owning team. Never create duplicate tickets. Build a performance changelog over time through consistent ticket updates.

README

# Slow Endpoint Finder Slow Endpoint Finder is a weekly performance-monitoring agent that scans your API endpoints in Honeycomb, calculates latency percentiles (p50, p95, p99) over the last 7 days, compares them to the prior 7-day window, and flags any endpoint whose p95 exceeds a configurable threshold (default 500 ms). For each flagged endpoint, it computes the p95 delta week-over-week, gathers request volume/day, and either updates an existing Linear ticket (tagged “performance”) with a new comment and Honeycomb query link, or creates a new ticket if none exists—avoiding duplicates and building a consistent performance changelog over time. The agent uses the following MCP servers/tools: Honeycomb MCP (https://mcp.honeycomb.io/mcp) to run queries and retrieve latency and volume; Linear MCP (https://mcp.linear.app/mcp) to search, create, and update tickets. It runs on the claude-sonnet-4-6 model and includes agent_toolset_20260401 plus mcp_toolset integrations for both Honeycomb and Linear with an always_allow permission policy. Tickets are tagged “performance” and assigned to the owning team as defined in your Linear setup. Quick start: - Prerequisites: Ensure network access and credentials for Honeycomb (API key, dataset/environment) and Linear (API key, team/project). Confirm the MCP endpoints above are reachable and configured. Optionally ensure there’s a clear endpoint-to-owning-team mapping in Linear. - Configure: Keep the default p95 threshold (500 ms) or adjust it in the agent configuration. Verify the time windows (last 7 days vs prior 7 days) match your reporting cadence. - Test run: Execute the agent once manually to verify it can query Honeycomb and read/write to Linear. Start with a higher p95 threshold to reduce noise during validation. - Schedule: Set a weekly schedule (for example, Monday morning) via your automation platform (cron, CI/CD, or orchestrator). Monitor logs and resulting Linear tickets/comments tagged “performance” to track progress and week-over-week changes.

MCP Servers

honeycomb
linear

Agent Configuration (YAML)

name: Slow Endpoint Finder
description: >-
Runs weekly to query Honeycomb for API endpoint latency (p50, p95, p99), flags endpoints exceeding a p95 threshold,
compares against last week, and creates or updates Linear tickets for each slow endpoint.
model: claude-sonnet-4-6
system: >-
You are a performance monitoring agent that runs weekly to identify slow API endpoints. Each run: query Honeycomb for
all API endpoints and their p50, p95, and p99 latency over the past 7 days, plus the prior 7-day window for
comparison. Flag any endpoint where p95 exceeds 500ms (configurable). For each flagged endpoint, compare current p95
against last week's p95 and note the delta. Check Linear for existing open tickets tagged "performance" for that
endpoint. If one exists, add a comment with updated p95, volume/day, and a Honeycomb query link. If not, create a new
ticket with: endpoint path, current p95, last week's p95, request volume per day, and Honeycomb link — tagged
"performance" and assigned to the owning team. Never create duplicate tickets. Build a performance changelog over time
through consistent ticket updates.
mcp_servers:
- name: honeycomb
url: https://mcp.honeycomb.io/mcp
type: url
- name: linear
url: https://mcp.linear.app/mcp
type: url
tools:
- type: agent_toolset_20260401
- type: mcp_toolset
mcp_server_name: honeycomb
default_config:
permission_policy:
type: always_allow
- type: mcp_toolset
mcp_server_name: linear
default_config:
permission_policy:
type: always_allow
skills: []