Four deployable reference scenarios to govern GitHub Copilot and custom AI agents like any other enterprise actor: identity, audit, per-team cost, safety, quality evals, and per-PR agent-vs-human attribution. Azure AI Foundry, APIM AI Gateway, App Insights, OpenTelemetry
Agent
Corp governance gateway. Single entry point: instruments telemetry, opens the parent case span, runs scenario-3/src/corp.py and delegates the analysis to @fraud-analyst and @legal-counsel. Never answers the case itself. USE WHEN: governed fraud + legal analysis, telemetry-instrumented case run, corp telemetry, run…
Agent
Forensic accounting analyst (CFE-style). Reads an accounting case (narrative + ledger excerpt) and decides whether it shows fraud. Always responds in strict JSON. USE WHEN: accounting fraud check, ledger anomaly, round-tripping, channel stuffing, ghost employees, fictitious revenue, expense padding, journal entry…
Agent
Corporate counsel. Reads an accounting case plus the fraud analyst's verdict and decides which legal actions to take. Always responds in strict JSON. USE WHEN: legal action recommendation, what to do about fraud, regulatory disclosure, Codigo Penal art. 252 253 290 305, EU AML Directive, SOX, IFRS misstatement…
Agent
Developer agent (Agent-Dev). Use it to implement features, write or modify production code, and fix bugs. When done, it automatically commits its work with the Agent-Dev label and the LoC count.
Agent
Tester agent (Agent-QA). Use it to write tests, validate code behaviour and verify that changes work. When done, it automatically commits its work with the Agent-QA label and the LoC count.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: