Skip to content
Back to skills

Incident Response

ASecurity

Use when responding to or forensically investigating an incident — triage acquisition (Velociraptor/KAPE), Volatility 3 memory forensics, Chainsaw/Hayabusa EVTX timelining, anti-forensics detection, cloud IR, ransomware/ESXi response

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 17, 2026
securitypythongoshellbashawsgcpazuregitapisecurity

Works with

  • cli
  • api

Security analysis

A100/100

Pro scans all 17 files and shows the line behind each finding

Scanned October 17, 2026

npx -y skills add NetVar1337/agent-skill-canon --skill incident-response --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Incident Response?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Incident Response
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/netvar1337-incident-response/badge)](https://www.skillsdirectory.com/skills/netvar1337-incident-response)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: incident-response
description: Use when responding to or forensically investigating an incident — triage acquisition (Velociraptor/KAPE), Volatility 3 memory forensics, Chainsaw/Hayabusa EVTX timelining, anti-forensics detection, cloud IR, ransomware/ESXi response
metadata:
  type: defensive
  phase: response
  tools: velociraptor, volatility3, chainsaw, hayabusa, plaso, timesketch, uac, MFTECmd, EvtxECmd, dissect, certutil, awscli, az, gcloud
  mitre: TA0005
kill_chain:
  phase: [report]
  step: [8]
  attck_tactics: [TA0005, TA0007, TA0040, TA0010]
  attck_techniques: [T1070, T1070.001, T1070.004, T1070.006, T1562.001, T1070.008, T1055, T1014, T1003.001, T1486, T1490, T1485, T1078.004, T1552.005, T1528, T1219]
depends_on: [red-team-ops, threat-hunting]
feeds_into: [threat-hunting, malware-analysis]
inputs: [memory_dumps, disk_images, triage_packages, log_data, cloud_audit_logs]
outputs: [timeline, ioc_list, forensic_report, containment_actions, root_cause]
references:
  - references/triage-collection.md
  - references/memory-forensics.md
  - references/windows-evtx-timeline.md
  - references/anti-forensics-detection.md
  - references/cloud-ir.md
  - references/ransomware-esxi-ir.md
  - references/repo-compromise-forensics.md
scripts:
  - scripts/triage_collector.py
  - scripts/vol3_triage.py
  - scripts/ebpf_rootkit_hunt.sh
  - scripts/evtx_hunt.sh
  - scripts/timestomp_detect.py
  - scripts/cloud_ir_collect.py
  - scripts/ransomware_triage.ps1
  - scripts/dangling_commit_finder.py
  - scripts/gharchive_recover.py
---

# Incident Response & Digital Forensics

## When to Activate

- Active security incident: triage, scoping, evidence acquisition, containment, eradication
- Memory forensics — process injection, rootkit (incl. eBPF), credential-theft, network artifacts
- Windows event-log / artifact timelining and super-timeline reconstruction
- Anti-forensics detection — timestomping, log clearing, secure deletion, VSS recovery
- Cloud incident response — AWS/Azure/GCP identity-plane attacks and forensic collection
- Ransomware / extortion response — hypervisor (ESXi) encryption, backup destruction, fast-dwell intrusions
- Verifying suspect DFIR tooling used as adversary persistence (Velociraptor CVE-2025-6264)
- Repository-compromise post-mortem — a poisoned public repo / force-pushed malicious commit / deleted PR (recover via dangling commits + GH Archive + Wayback + Events API; see `references/repo-compromise-forensics.md`)

## Technique Map

| Technique | ATT&CK | CWE | Reference | Script |
|-----------|--------|-----|-----------|--------|
| Order-of-volatility live triage (Velociraptor/KAPE/UAC/CatScale) | T1074 | CWE-778 | references/triage-collection.md | scripts/triage_collector.py |
| Offline collector build + RAM acquisition (winpmem/LiME/AVML) | T1074 | CWE-778 | references/triage-collection.md | scripts/triage_collector.py |
| Suspect-tooling verification (Velociraptor CVE-2025-6264) | T1219 | CWE-269 | references/triage-collection.md | scripts/triage_collector.py |
| Volatility 3 process/injection analysis (malfind, hollow) | T1055 | CWE-noinfo | references/memory-forensics.md | scripts/vol3_triage.py |
| Credential extraction from memory (LSASS, hives) | T1003.001 | CWE-522 | references/memory-forensics.md | scripts/vol3_triage.py |
| Kernel + eBPF rootkit detection (LinkPro, linux.ebpf) | T1014 | CWE-269 | references/memory-forensics.md | scripts/ebpf_rootkit_hunt.sh |
| EVTX Sigma hunting & fast timeline (Chainsaw/Hayabusa) | T1070.001 | CWE-778 | references/windows-evtx-timeline.md | scripts/evtx_hunt.sh |
| Super-timeline (plaso) + Timesketch correlation | T1070 | CWE-778 | references/windows-evtx-timeline.md | scripts/evtx_hunt.sh |
| Timestomping detection ($SI vs $FN, USN FILE_CREATE) | T1070.006 | CWE-noinfo | references/anti-forensics-detection.md | scripts/timestomp_detect.py |
| Log/journal clearing & VSS recovery | T1070.001, T1490 | CWE-778 | references/anti-forensics-detection.md | scripts/timestomp_detect.py |
| Cloud IR — IMDSv2/SSRF cred theft, CloudTrail/GuardDuty | T1552.005, T1078.004 | CWE-918 | references/cloud-ir.md | scripts/cloud_ir_collect.py |
| Entra ID / token theft, identity-plane containment | T1528, T1078.004 | CWE-287 | references/cloud-ir.md | scripts/cloud_ir_collect.py |
| Ransomware rapid triage (Windows/Linux) | T1486, T1490, T1485 | CWE-noinfo | references/ransomware-esxi-ir.md | scripts/ransomware_triage.ps1 |
| ESXi / hypervisor ransomware response (UNC3944) | T1486 | CWE-noinfo | references/ransomware-esxi-ir.md | scripts/ransomware_triage.ps1 |

## Quick Start

```bash
# 0. PRESERVE ORDER OF VOLATILITY — RAM before disk, never reboot a live host first.
#    Windows RAM:  winpmem_mini_x64.exe mem.raw        Linux RAM: AVML  ./avml mem.lime
# 1. Network-wide / endpoint triage (pick one):
python3 scripts/triage_collector.py --os auto --out /evidence --velociraptor-collector
#    Verify any Velociraptor already on-host is NOT adversary persistence (CVE-2025-6264):
python3 scripts/triage_collector.py --check-velociraptor   # flags <0.73.5 / unknown service

# 2. Memory forensics (Windows or Linux dump):
python3 scripts/vol3_triage.py -f /evidence/mem.raw --os windows --hunt-injection --dump-suspect
bash   scripts/ebpf_rootkit_hunt.sh   # Linux live/IR eBPF rootkit hunt (LinkPro-aware)

# 3. Windows event-log fast timeline + Sigma hunt:
bash scripts/evtx_hunt.sh -d /evidence/C/Windows/System32/winevt/Logs -o /evidence/timeline

# 4. Anti-forensics: timestomp / USN tamper detection from $MFT + $J:
python3 scripts/timestomp_detect.py --mft /evidence/mft.csv --usn /evidence/usn.csv

# 5. Cloud breach (identity-plane first):
python3 scripts/cloud_ir_collect.py aws --collect-cloudtrail --contain-key AKIA... --enforce-imdsv2

# 6. Ransomware on a Windows host (rapid scope, do BEFORE eradication):
powershell -ep bypass -File scripts/ransomware_triage.ps1 -OutDir C:\IR
```

## OPSEC & Detection (summary)

> IR is defensive; "OPSEC" below = handling rules that keep evidence admissible and avoid tipping off an adversary who may be monitoring (UNC3944 joins IR bridges in real time).

| Technique | Telemetry / IOC | Detection (Sigma / EDR) | OPSEC / evidence note |
|-----------|-----------------|--------------------------|------------------------|
| Live triage | New service/scheduled task for collector; large file writes to evidence path | Baseline expected DFIR tooling; alert on unsigned collectors | Collect RAM first; never write evidence to the suspect volume; hash everything |
| Velociraptor abuse | velociraptor.exe svc <0.73.5; MSI from Azure Blob; relaunch after isolation | Sigma: unexpected Velociraptor service install; CVE-2025-6264 UpdateConfig | Treat unexpected Velociraptor as persistence, not your tooling |
| Memory injection | RX/RWX private VAD not file-backed (malfind); reparented svchost | Vol3 malfind/hollowprocesses; EDR unbacked-exec | Document plugin+offset provenance; keep raw dump read-only |
| eBPF rootkit | bpf_override_return; getdents/sys_bpf hooks; XDP magic-packet (win=54321); /etc/ld.so.preload | linux.ebpf vs baseline; ss(netlink) vs /proc/net diff; YARA MAL_LinkPro_* | bpftool/ps/ss lie on host — acquire RAM out-of-band (hypervisor/LiME RO) |
| EVTX clearing | 1102 (Security cleared), 104 (System cleared), gaps in EventRecordID | Chainsaw/Hayabusa Sigma; alert on 1102/104 | Pull EVTX from VSS/disk image, not the tampered live log |
| Timestomp | $SI ≠ $FN create time; sub-second zeros; USN FILE_CREATE mismatch | timestomp_detect.py; MFTECmd Created0x10 vs Created0x30 | $FN is harder to forge — anchor truth to it + USN/$LogFile |
| Cloud cred theft | InstanceCredentialExfiltration.OutsideAWS; impossible-travel sign-in; CloudTrail StopLogging | GuardDuty findings; Sentinel KQL risky sign-ins | Snapshot+immutable-export BEFORE remediation; logs to a SIEM the attacker can't reach |
| Ransomware/ESXi | Mass file rename/ext change; vCenter/ESXi SSH on; bulk vim-cmd VM power-off | SIEM: high-volume VM power-off from one host; vpxuser anomalies | Image before decrypt attempts; preserve note + sample; assume comms compromised |

## Deep Dives

- **references/triage-collection.md** — Order of volatility, RAM acquisition (winpmem/AVML/LiME), Velociraptor 0.75 offline collectors & hunts, KAPE targets, UAC/CatScale for Unix/ESXi, suspect-tooling verification incl. Velociraptor CVE-2025-6264.
- **references/memory-forensics.md** — Volatility 3 symbol-table workflow, Windows injection/credential/rootkit plugins, Linux pslist/check_syscall/hidden_modules, eBPF rootkit detection (LinkPro, linux.ebpf, bpf_override_return), dump extraction.
- **references/windows-evtx-timeline.md** — Chainsaw v2 & Hayabusa v3 Sigma hunting, key Event IDs, EvtxECmd/Eric Zimmerman parsers, plaso super-timeline, Timesketch + Dissect/Acquire enterprise scaling.
- **references/anti-forensics-detection.md** — Timestomping ($SI/$FN + USN cross-validation), log/journal clearing (1102/104, $LogFile gaps), secure-deletion artifacts, VSS recovery, $MFT/$J/$LogFile QuadLink correlation.
- **references/cloud-ir.md** — NIST SP 800-61r3 / SP 800-201, AWS CloudTrail/GuardDuty/IMDSv2-SSRF, Azure Entra ID token theft & KQL, GCP audit logs, identity-plane containment, automated evidence preservation.
- **references/ransomware-esxi-ir.md** — Ransomware rapid triage & decision flow, Scattered Spider/UNC3944 ESXi LotL chain, backup destruction, cross-platform builders (LockBit 5.0/DragonForce), containment & negotiation hygiene.
- **references/repo-compromise-forensics.md** — poisoned public-repo post-mortem from four INDEPENDENT, attacker-uncontrollable sources: dangling/force-pushed commits (`dangling_commit_finder.py`, git fsck via `git_safe`), GH Archive + Wayback CDX + live Events API (`gharchive_recover.py`); hypothesis→verify-at-source (evidence_kit)→adversarial-check→report, attribution-with-confidence, BigQuery kept optional.

Files in this skill

  • SKILL.md9.8 KB
  • references/anti-forensics-detection.md7.1 KB
  • references/cloud-ir.md8.1 KB
  • references/memory-forensics.md8.5 KB
  • references/ransomware-esxi-ir.md7.7 KB
  • references/repo-compromise-forensics.md3.9 KB
  • references/triage-collection.md9.3 KB
  • references/windows-evtx-timeline.md6.9 KB
  • scripts/cloud_ir_collect.py8.4 KB
  • scripts/dangling_commit_finder.py6.6 KB
  • scripts/ebpf_rootkit_hunt.sh6.5 KB
  • scripts/evtx_hunt.sh5.6 KB
  • scripts/gharchive_recover.py6.5 KB
  • scripts/ransomware_triage.ps17.3 KB
  • scripts/timestomp_detect.py6.6 KB
  • scripts/triage_collector.py10.1 KB
  • scripts/vol3_triage.py8.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.