aozyildirim/agena/integration-rule-engine

Integration Rule Engine

When a Sentry / NewRelic / Jira / Azure DevOps event lands, you want to auto-tag it, route it to the right AI reviewer agent, and bump priority based on attributes — without rebuilding the routing logic in code per source. The Integration Rule Engine collapses all that into one rule table.

How to apply this pattern

1. Single rule table with provider + match (JSON) +

action (JSON) + priority (rule precedence) + is_active. Same shape across providers; the matcher just checks the appropriate attributes.

1. Match clause — matches a single field with one operator. Common

operators:

1. Action clause — what happens when the rule matches:

• tag: add a tag string to the imported task (e.g. security_review)
• preferred_agent_role: which reviewer agent to route through

(e.g. security_developer)

• priority: override priority (critical, high, etc.)
• target_repo: override the default repo mapping for this task

1. Evaluate at import time, not at agent run time. Stamp the action

into the task's metadata so downstream agents and the UI see the same routing without re-evaluating.

1. First-match-wins with explicit priority. A rule with higher

priority gets evaluated first; a single rule wins per import.

Example rules

# Security tickets reported by the security team go to the
# OWASP-aware reviewer with critical priority
- provider: jira
  match: { field: reporter, op: in, values: [security@yourco.com] }
  action:
    tag: security_review
    preferred_agent_role: security_developer
    priority: critical
 
# Sentry errors in the payments project always route to a senior
# backend reviewer
- provider: sentry
  match: { field: project, op: equals, value: payments }
  action:
    tag: payments
    preferred_agent_role: senior_backend
    priority: high
 
# Azure DevOps work items in a specific area path
- provider: azure_devops
  match: { field: area_path, op: prefix, value: "Project\\Frontend" }
  action:
    tag: frontend
    preferred_agent_role: a11y_reviewer

Notes

• Declarative > imperative. Don't sprinkle if reporter == X

branches across import code. The whole point is letting non-engineers edit routes from a UI.

• Same engine for every provider. Sentry, NewRelic, Jira, Azure

DevOps imports all flow through the same rule evaluator. Adding a new provider becomes "expose its attributes to the matcher" instead of "rewrite routing".

• Audit the route. Every applied rule should leave a trail on the

task: which rule fired, which fields matched. Important when a reviewer asks "why did this end up with security_review?"

• `Preferred Agent Role` stamped on description metadata is a clean

hand-off to the review service. The reviewer-resolution code then honours that without needing to re-evaluate the rule.

• Test rules before publishing: provide a "dry run" mode where the

rule editor shows which past tickets / errors would have matched. Catches greedy regex disasters before they fire on real imports.