v1.0 — Agent-Driven Security Operations

Enable agentic remediation
across your AI tooling stack.

A curated, community-driven library of recipes that turn the AI assistants your engineers already use into autonomous remediators — patching vulnerabilities, fixing flaky tests, and closing security findings without leaving the IDE.

Quick Start 5 min Start with the primer
The AI tools already in your stack Curated, practitioner-tested recipes Community-maintained - contribution welcomed!

Pick your agent

Click any tool to see the step-by-step recipe for enabling agentic remediation in your environment.

GitHub Copilot
copilot.github.com
View recipe
Devin
cognition.dev
View recipe
Cursor
cursor.com
View recipe
Codex
openai.com/codex
View recipe
Claude
claude.com / claude code
View recipe
Community-Driven · Reviewed

Prompt Library

Share the prompts, rules, skills, and instruction files that are actually shipping fixes on your team. Stop re-inventing copilot-instructions.md — borrow from the people who've already iterated theirs. Contribute yours back so the next team starts further along than you did.

copilot-instructions.md
CLAUDE.md · skills
.cursor/rules/*.mdc
AGENTS.md
Devin Knowledge entries
PreToolUse hooks
Automation · Deterministic

Automation, not agentic

Before you reach for a model, reach for a --fix flag. Dependabot, Renovate, npm audit fix, pip-audit, go mod tidy, and GitHub code scanning do a huge amount of risk-reduction work without an LLM in the loop — and they pair well with the agentic flows on this site. Use deterministic tools where you can; save the agent for the places deterministic tools can't reach.

dependabot renovate npm audit pip-audit code scanning
Click to find out more
Security-operated · Reviewer-gated

Agentic workflows a security team operates

Reference workflows a security team operates on engineering's behalf — not ones engineering teams are asked to run. Bounded scope, reversible output (always a PR, never a merge), measurable outcomes, and clean failure modes. The output is PRs tagged with an auto-remediation label (rename to your org's convention), or triage tickets when the agent stops cleanly.

sensitive data elements vulnerable dependencies
Click to find out more
Context Layer · Scoped

MCP Server access

An agent is only as fast as the context it can reach. The more of your risk-relevant signals — findings, ownership, tickets, runbooks, build status — are exposed through MCP with the right scopes, the shorter the distance from a new finding to a reviewed PR. Scoped tokens, typed interfaces, rate limits, audit logs: Security stays in control while Engineering ships faster.

scoped tokens typed tools audit logs rate limits per-agent wiring
Click to find out more