More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Threat Hunting and Detection Engineering

    1

    Hypothesis-led threat hunting: baseline normal in SIEM/EDR data, stack queries into a funnel, convert confirmed hunts into detection rules with a false-positive budget, and validate in the lab.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Threat Hunting and Detection Engineering

    Threat Hunting and Detection Engineering

    Example session with this skill installed

    Over the last month our EDR flagged a few workstations running rundll32 with no command line at odd hours. We have 90 days of endpoint process logs and authentication logs in our SIEM, and a small SOC team. Can you run this as a hunt and show us what you would actually do, step by step?

    • Read your context and instructions
    • Compiled the threat hunting

    HUNT — rundll32 with empty command lines, off-hours.

    Hypothesis (scored 7/8): an intruder is proxying execution through rundll32 to blend with OS noise. Data: process-creation logs record full command lines and parents; auth logs cover the same 90 days. Success = one chain linking an empty-command-line rundll32 to a suspicious parent or an unusual logon on the same host. Negative = every instance traces to a signed OS or vendor parent inside its normal pattern.

    Baseline: rundll32 fires ~1,900 times/day fleet-wide. Legitimate parents are dominated by the OS shell, the print spooler, and two vendor updaters; empty command lines are normal for one updater's maintenance task, which runs 01:00–03:00 under a service account. That carve-out is the noise floor — exclude it by parent hash and account, not by time alone.

    Funnel: broad pull of empty-command-line rundll32 over 90 days: 412 events. Filter out the known updater and shell parents: 37 remain. Correlate against parents outside the baseline and logons within 30 minutes: 4 events on 2 hosts share one parent — a user-writable sync tool — and one host shows a network logon from an account that never touches it. Timeline on that host: sync tool starts, rundll32 (empty) follows, outbound connection 6 minutes later. That narrative fits execution-then-beacon; the other 3 events resolve to a misconfigured printer driver.

    Verdict: escalate the one host as a package (timeline, IOCs, scope note). Convert: a Sigma-style rule — rundll32 with empty or unusual command line AND parent outside the approved set — backfilled over the 90 days at ~3 hits/day, inside budget with the updater allow-listed. Atomic lab test scheduled before the rule is enabled, and the negative coverage for the rest of the fleet is written into the hunt log.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Alert queues only show you what your detectors already know. This skill runs threat hunting as a discipline: every hunt starts as a falsifiable hypothesis about adversary behavior — scored on a rubric before query time is spent — and normal gets baselined before anyone anomaly-hunts, so the noise floor stops eating the schedule. Queries run as a stacked funnel (broad, filter, correlate, timeline) that shrinks candidates by roughly an order of magnitude per stage. Confirmed techniques convert into detection rules written as portable code with explicit field mappings, each carrying a false-positive budget estimated from your own data and a decision table for when a rule should alert, aggregate, or stay a hunt query. Nothing guards production unproven: new rules pass atomic lab tests and historical backfill first, tuned from the backfill evidence. Negatives are logged as coverage, positives escalate as timeline and IOC packages rather than chat messages, and every hunt feeds a prioritized backlog of follow-up hypotheses.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.

    • It cannot manufacture telemetry: if your logging doesn't capture a behavior, the honest output is a data gap and a reformulated hypothesis, not a verdict.
    • Validation assumes you have (or can build) an isolated lab to run adversary techniques safely; without one, rules ship with an explicit unvalidated flag.
    • It hands confirmed intrusions to incident response with an evidence package — containment and eradication are IR's process, not this skill's.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 24 days ago

    Permissions

    No special permissions declared or detected

    Listed24 days ago
    Updated21 days ago

    What's inside

    Frequently Asked Questions

    Alerts tell you what existing detections already caught. Hunting starts from a hypothesis about what they might be missing, tests it against baselined data, and converts anything confirmed into a new detection — so the queue itself gets better every cycle.

    A logged negative: the technique was tested for, the data was sufficient, and the environment showed none of it. That documents coverage, stops the next analyst re-hunting the same ground, and sharpens the backlog of what to hunt next.

    No. The method is platform-agnostic; rules are written in Sigma or your platform's native format with the field mapping made explicit. What you do need is process and authentication telemetry you can query over a meaningful window.

    New in Security