.png)
Falco Feeds extends the power of Falco by giving open source-focused companies access to expert-written rules that are continuously updated as new threats are discovered.
Detection is just the start
When Sysdig fires a high-severity runtime alert (e.g., a cryptominer running inside a web pod, a reverse shell connecting outbound, or credentials being pulled from the cloud metadata endpoint), most teams know what they're looking at within seconds. What they don't know is what to do about it.
What depends on this workload? If I kill the container, does the sidecar mesh break? Does a StatefulSet start throwing errors? Is there a PodDisruptionBudget that blocks me entirely? Did I just destroy evidence before I could collect it?
Today, answering those questions under pressure requires pulling together an SRE, a security engineer, and a cloud admin. Usually in a Slack incident channel. Usually at 2am. Usually racing against an attacker who isn't slowing down.
This is the problem the new Runtime Remediation Skill solves.
Here, headless cloud security changes the response model. Instead of forcing analysts to pivot from an alert into dashboards, runbooks, cloud consoles, and chat threads, Sysdig brings runtime intelligence and governed response actions into the operational workflows where teams already work. The result is not autonomous remediation for its own sake; it is faster, safer response with the analyst in control at every step.
Response is a knowledge problem, not just a tooling problem
Anyone who has worked a live cloud incident knows the triage phase rarely takes long. It's the next five minutes that hurt: determining blast radius, coordinating the right people, and deciding whether to kill a container or isolate it or grab a capture first, all without breaking something else in the process.
These judgments are usually tribal knowledge. This know-how lives in the heads of your most experienced engineers. It doesn't scale, it doesn't document itself, and it absolutely doesn't show up at 3am when those engineers are asleep.
The Runtime Remediation Skill encodes that judgment and runs it directly in the analyst's terminal, with the analyst's hand on the wheel at every step.
What the Runtime Remediation Skill actually does
The Runtime Remediation Skill is an agent skill: a set of instructions that tells your AI coding agent how to execute a specific security workflow, step by step, using the tools and data it already has access to. It does not ask an AI agent to improvise remediation from generic guidance. It gives the agent a governed response workflow grounded in Sysdig's runtime intelligence and cloud-native security expertise — one that reads the blast radius, proposes ordered actions, waits for your confirmation on every destructive step, and watches for respawn after containment.
The skill runs a six-step response flow, fully narrated so the analyst always sees what's about to happen before it happens. It starts by building a real-time blast-radius map of the affected workload: service dependencies, sidecars, ServiceAccount permissions, cloud identity bindings, and operational constraints like PodDisruptionBudgets. This map is read-only, before anything is touched, with no guessing about what breaks if you act.
From there the skill proposes an ordered set of response actions. For every action, it shows exactly what the action does, what it breaks, whether it can be undone, and the precise undo call if it can. Every destructive action requires its own explicit confirmation. There is no "approve all." That's by design.
This follows the core headless security model: the agent proposes, the human decides, and every action is logged.
The action set spans the full response surface: forensic collection (acquire the binary, capture syscalls, and snapshot the volume) before containment, network isolation and process termination for active threats, and IAM session revocation for credential theft. All this occurs in the right order, not just the obvious one. When credentials have been stolen, the attacker has them in hand, rather than in the pod. Revoking the IAM session first is the correct move. The skill knows that.
After acting, it watches for five minutes to confirm the threat didn't respawn. It closes as cleared, still_active, or inconclusive and says so plainly, instead of leaving the analyst to wonder. The full audit trail, with UTC timestamps, action decisions, and watcher results, appends to the incident ticket automatically.
This skill runs through the Sysdig MCP server, which gives your AI agent structured access to Sysdig's runtime detections, workload context, and response actions. It requires the MCP server to be registered via OAuth.
The bigger shift this represents
Following the Runtime Investigation Skill, the Runtime Remediation Skill closes the loop: from detection to scoped investigation to safe and auditable response, all inside the tools where teams already work.
This is what headless cloud security looks like in practice. It’s not a new dashboard to context-switch into, or a response playbook that lives in a wiki nobody reads during an incident. The teams that detect threats fastest have always had an edge. Now they can contain them faster too — without the tribal knowledge bottleneck, without the 2am escalation chain, and without guessing what breaks next.
The Runtime Remediation Skill is available in Public Beta today via Claude Code and any Agent Skills-compatible environment.
Request a demo to see the full detection-to-response loop in action.