TechPrufer
Get started
Figma
Figma logoFigma · Feature trail

Security Alert Triage Agents

3 posts · Jul 2026

What this trail covers

Figma is building an agentic system to automate security alert triage. This system leverages retrieval-augmented generation (RAG) to provide historical context for alerts and then uses specialized agents with scoped toolkits to investigate and resolve issues. The system aims to reduce on-call burden and improve the speed and accuracy of security incident response. This post details the evolution of the security review process, focusing on improving precision and recall through agentic systems, a shared policy, and an adjudicator. The system now includes provider failover, retry policies, and detailed telemetry. The prompt has been ablated, and a TypeScript service has been built to host the reviewer. The system now measures recall with a corpus of 66 real vulnerabilities and aims for 70% precision.

All Figma trails
  1. How we secure Figma’s internal systems with agentsThis post details the evolution of Figma's security alert triage system, moving from a manual, context-gathering process to a fully agentic system. Key technical contributions include the development of a RAG layer using AWS Bedrock Knowledge Bases and Amazon Kendra for surfacing historical alert context, including indexing structured alert fields and user-contributed investigation context from Slack. It also describes the addition of an agentic layer built with Tines, featuring intent routing to specialized agents with scoped toolkits (e.g., Okta, North Pole Security Workshop, Wiz) for automated investigation, code change generation, and PR creation. The post highlights metrics like a 20% drop in on-call pages due to automated severity downgrading and discusses the importance of auditability and scoped tool access for agents.
  2. How Figma stays ahead of vulnerabilities with agentsThis post details the evolution of Figma's agentic security review process, focusing on improving precision and recall. Key contributions include: the development of a shared policy that powers AI checks during code generation, pull request review, and auditing; the implementation of a universal, self-serve, and bi-directionally instrumented PR review loop with author feedback for precision and replaying known-bad commits for recall; the use of multiple LLMs (Claude Code, Codex) for broader coverage; the establishment of a 70% precision goal and the process to achieve it through hand-labeling and policy refinement; the realization that the policy itself serves as a comprehensive threat model; the migration of the reviewer into a maintainable TypeScript service with provider failover and retry policies; the ablation of the prompt to leverage newer LLM capabilities; and the addition of an adjudicator to re-examine borderline dismissals, improving recall by ~30%. The post also outlines the measurement of recall using a corpus of 66 real vulnerabilities.
  3. Got skills? Make the Figma agent a better collaboratorIntroduces 'skills' as a mechanism to teach the Figma agent reusable workflows and team-specific expertise. Details how skills can be used for simulating feedback, upholding standards, and providing new user perspectives. Explains how skills integrate with external tools via MCP connectors for logging feedback, incorporating research, and backing design decisions with data.