aeondave/forensic-technique

Forensic technique

Goal: produce defensible, reproducible findings from heterogeneous evidence.

When this technique applies

• Potential compromise requiring reconstruction of user/system activity.
• Evidence includes one or more of: disk image, ISO media, RAM dump, PCAP.
• Need to correlate endpoint, filesystem, and network timelines.
• Need report-ready output for incident response, legal, or internal review.

Boundary with offensive-tools

This skill explains methodology and decision flow. Tool-specific syntax belongs to offensive-tools/* skills.

Initial triage

Before extracting deeply, classify the evidence mix and decide what source can answer the objective fastest with the least interpretive risk.

• Starting state: is the case driven by disk, memory, PCAP, ISO/media, logs, or a mixed evidence set?
• First questions: what is the investigation objective, which artifacts are most volatile or most probative, and what timeline window and provenance constraints apply?
• Immediate actions: verify integrity, normalize time and scope, then inspect the highest-value source first instead of carving everything.
• Tool-family direction: use forensic acquisition/timeline families (sleuth-kit, autopsy, ftk-imager) for disk/media, volatility3 for memory, and zeek/tcpdump/wireshark for network evidence; use yara and capa only after suspicious artifacts are isolated.
• Escalation rule: require corroboration across sources before claiming compromise, execution, persistence, or exfiltration.

Agent operating model

The agent should keep this loop:

1. Preserve and scope evidence.
2. Prioritize by volatility and investigative value.
3. Examine source-specific artifacts.
4. Correlate across sources into one timeline.
5. Validate claims with independent corroboration.
6. Report facts, interpretation boundaries, and recommendations.

Do not move to deep extraction/carving before timeline framing and hypothesis definition.

Core forensic model

1. Preserve: isolate evidence, hash, and record provenance needed for reproducibility.
2. Collect: acquire additional required artifacts with minimal alteration.
3. Examine: parse artifacts by source type (disk, memory, network).
4. Correlate: build a unified timeline across sources.
5. Conclude: state what is supported by evidence vs assumptions.
6. Report: produce concise, reproducible findings and follow-up actions.

Evidence-first prioritization

Use volatility and value to decide acquisition/examination order.

1. Live volatile artifacts (RAM, active connections, running processes)
2. Network evidence at risk of rotation (live captures, ephemeral logs)
3. Endpoint high-value non-volatile artifacts (event logs, browser data, registry)
4. Full disk image deep examination
5. Enrichment sources (threat intel, external context)

Objective-driven workflows

1. Disk image investigation (E01/DD/RAW)

Use when you need file-system artifacts, deleted-content recovery, execution traces, and persistence evidence.

• Start with integrity + partition mapping.
• Enumerate file systems and prioritize user/profile, startup, logs, scheduled tasks, shell history.
• On Windows evidence, prioritize execution artifacts such as Prefetch, LNK, Jump Lists, AmCache, ShimCache, SRUM, PowerShell logs, services, and scheduled tasks.
• Build an initial timeline before deep carving.
• Recover/decode only artifacts linked to hypotheses (do not carve everything by default).
• Validate notable findings in at least one secondary source (e.g., link file + event log + registry).

Primary tool families:

• offensive-tools/forensic/sleuth-kit/
• offensive-tools/forensic/autopsy/
• offensive-tools/forensic/ftk-imager/
• offensive-tools/forensic/yara/
• offensive-tools/forensic/capa/ — capability detection on extracted executables (identifies malware families, TTPs, embedded shellcode)

2. ISO media investigation

Use when the evidence is installer-like media, archives, or mounted image content.

• Verify hash and mount read-only.
• Inventory executable/script content and autorun-related metadata.
• Extract Office/PDF/archive payload chains, embedded objects, external template references, shortcut chains, and script launchers before any dynamic handling.
• Compare bundle contents against expected vendor structure and signatures.
• Extract suspicious binaries/configs for static triage and rule scanning.
• Correlate ISO-delivered artifacts with host execution traces from endpoint evidence.

Primary tool families:

• offensive-tools/forensic/sleuth-kit/
• offensive-tools/rev/binwalk/
• offensive-tools/forensic/yara/

3. PCAP/network forensics

Use when reconstructing communication, exfiltration paths, C2 patterns, or lateral movement.

• Build traffic summary first (top talkers, protocols, unusual destinations).
• Pivot by DNS, TLS metadata, HTTP objects, and transfer channels.
• Identify sequencing: initial access indicators, staging, beaconing, bulk transfer.
• Tie network sessions to endpoint process/user context when possible.

Primary tool families:

• offensive-tools/forensic/zeek/
• offensive-tools/forensic/tcpdump/
• offensive-tools/network/wireshark/

4. Memory-centered investigation

Use when malware is fileless/injected, or when disk evidence is incomplete.

• Enumerate processes/parents, command lines, sockets, loaded modules.
• Hunt for injected regions and anomalous memory protections.
• Extract process artifacts and correlate with on-disk binaries and PCAP events.
• Treat memory findings as high-confidence for “what was running now,” then confirm persistence on disk.

Primary tool families:

• offensive-tools/forensic/volatility3/
• offensive-tools/forensic/yara/
• offensive-tools/forensic/capa/ — classify extracted process dumps or unpacked binaries against malware capability ruleset

5. Mixed-source incident reconstruction

Use when you have disk + PCAP + memory and need a single chronology.

• Create source-local timelines first.
• Normalize to a single timezone and clock offset model.
• Merge into one chronology with confidence tags:
  • Direct evidence (observed in source)
  • Corroborated evidence (confirmed by second source)
  • Inference (plausible but unconfirmed)
• Explicitly record competing explanations for ambiguous events.

6. Log-centric and objective-driven investigations

Use when evidence is mostly EVTX, registry hives, MFT snapshots, API trace logs, or objective-driven forensic tasks.

• Build an objective-to-artifact map first (which artifact can answer each objective decisively).
• Prefer deterministic extraction over broad hunting (exact key, exact event id, exact stream, exact record).
• For Windows log-heavy cases, prioritize: PowerShell ScriptBlock (4104), process creation (4688/Sysmon 1), service/task creation, Defender/AV alerts, and relevant registry keys. When an ETW trace (.etl) is provided, treat provider name + event id as the primary pointer (e.g. Microsoft-Windows-PowerShell, Microsoft-Antimalware-Scan-Interface, Microsoft-Windows-Kernel-Process).
• For Linux hosts, prioritize journalctl output from /var/log/journal/* (auth, sshd, sudo, systemd unit start/stop) and /var/log/audit/audit.log (auditd) for structured process/command evidence.
• For macOS hosts, use the Unified Log system (log collect, log show --predicate ..., or an .logarchive bundle) for process, network, and security signals; classic /var/log text logs are largely deprecated.
• For API trace cases, reconstruct sequence by API dependency chain (enumeration → allocation/write → execution).
• Mark each conclusion as: direct artifact fact vs inferred interpretation.

Primary tool families:

• offensive-tools/forensic/chainsaw/ — fast triage across EVTX, $MFT, ShimCache, SRUM, and registry hives with Sigma and built-in detection rules; produces multi-source execution timelines

Practical quality gates

• Provenance and hash records exist for all primary evidence.
• At least one key claim is corroborated across two independent sources.
• Timeline includes both malicious and benign context to avoid narrative bias.
• Every IOC or behavioral claim has exact source pointer (artifact + timestamp).
• Report distinguishes fact, interpretation, and recommendation.

Anti-patterns

• Starting with deep carving before timeline and hypothesis framing.
• Treating one artifact as conclusive without corroboration.
• Mixing local-time and UTC evidence without explicit normalization.
• Writing conclusion-first reports and backfilling evidence.

Required deliverables from the agent

1. Evidence inventory with integrity state.
2. Source-specific findings with exact pointers.
3. Correlated timeline with confidence labels.
4. High-confidence claims vs inference boundary.
5. Report-ready conclusions and next actions.

Resources

• references/evidence-preservation.md — Load for provenance, hashing, read-only workflow, and evidence-handling order.
• references/disk-and-iso-analysis.md — Load for filesystem/media triage, startup paths, selective extraction, and validation.
• references/windows-execution-indicators.md — Load for Prefetch, LNK, Jump Lists, AmCache, ShimCache, SRUM, and related execution artifacts.
• references/document-malware-analysis.md — Load for Office, PDF, archive, embedded-object, and external-template payload chains.
• references/pcap-analysis.md — Load for protocol triage, session reconstruction, transfer paths, and network chronology.
• references/memory-analysis.md — Load for process/module/socket triage, injected regions, transient secrets, and disk/network correlation.
• references/timeline-correlation.md — Load for clock normalization and merging source-local timelines with confidence labels.
• references/lessons-learned-patterns.md — Load for mixed-artifact heuristics and fast-answer investigation patterns.
• scripts/forensic_triage.py — Run to classify an evidence file and get the first forensic workflow/tool-family recommendation.