Computer-Use Agent Taxonomy中文
R03.05.01.04 · TASK CLASS

Disk, memory & binary forensics

Recover artifacts and extract indicators with forensic and reverse-engineering GUIs.

20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.

Typical applications
Autopsy · Ghidra · Volatility · Wireshark
Class GUI-need label
HY · Hybrid

Task proposals · 20

Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.

R03.05.01.04-T01 Carve recoverable images from a disk image

Original instruction

In Autopsy, add fixtures/for/case01.E01 as a data source and run file carving. Report how many recoverable JPEG/PNG files were carved and the sha256 of the largest one. Tag the carved files in the case.

Platforms
win · Windows lin · Linux
Applications
Autopsy
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click nav · Pan/zoom/scroll vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.json count matches the lab ground truth; the reported sha256 matches the largest carved image; the carved set is tagged in the Autopsy case db. GUI carving is graded via the case; a photorec CLI route is allowed to cross-check but tags must exist in the case.

Proposed partial credit

40% correct carved count; 30% correct largest-file hash; 30% files tagged in case

R03.05.01.04-T02 Build a file-activity timeline from a disk image

Original instruction

From fixtures/for/case01.E01, generate a timeline and identify the 3 files created or modified within the 10-minute window around the incident time in incident.txt. Export those timeline rows and tag the files.

Platforms
win · Windows lin · Linux
Applications
Autopsy
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product
WP10 · Annotations
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

The 3 reported file paths match the lab key for the incident window; exported timeline.csv rows carry correct MACB times; files tagged in the case. A fls/mactime CLI cross-check is allowed; final answer graded on the exported rows.

Proposed partial credit

50% three correct files in window; 30% correct MACB times; 20% exported+tagged

R03.05.01.04-T03 Recover deleted browser history

Original instruction

The user cleared browsing history on fixtures/for/laptop.E01. Recover the deleted SQLite history records and report the last 5 visited URLs with timestamps. Save them to history.csv and add a case note.

Platforms
win · Windows
Applications
Autopsy
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
HY · Hybrid
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

history.csv lists the 5 correct URLs+timestamps from the recovered records matching lab ground truth (order by time). Graded on the exported CSV; sqlite recovery via CLI is permitted as a cross-check.

Proposed partial credit

60% five correct URLs; 20% correct timestamps; 20% case note recorded

R03.05.01.04-T04 Extract malware config in Ghidra to a schema

Original instruction

In Ghidra, analyze the inert lab sample fixtures/for/sample.bin (static only, sandboxed). Locate the embedded configuration (C2 host, port, campaign id) and write malware_config.json per fixtures/for/schema.json. Annotate the config-decoding function in Ghidra.

Platforms
win · Windows lin · Linux
Applications
Ghidra
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment code · Code & commands
Work product
WP10 · Annotations
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

malware_config.json validates against the schema and its field values match the lab ground truth; the decoding function is renamed/commented in the Ghidra project. Static analysis only; sample never executed.

Proposed partial credit

30% config region located; 40% correct field values in schema; 30% function annotated in Ghidra

R03.05.01.04-T05 Annotate a decryption routine in Ghidra

Original instruction

The lab binary fixtures/for/stringhide.bin decrypts its strings at runtime. In Ghidra, identify the decryption routine, recover the key/algorithm statically, label the function and key variable, and submit the decrypted banner string to answer.txt.

Platforms
win · Windows lin · Linux
Applications
Ghidra
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment code · Code & commands
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

answer.txt equals the known decrypted banner; the decryption function and key variable are renamed in the Ghidra project; a decode script (if used) reproduces the string. Static only.

Proposed partial credit

40% routine located+labeled; 30% key/algorithm recovered; 30% correct decrypted string

R03.05.01.04-T06 Find the suspicious process in a memory dump

Original instruction

Using Volatility on fixtures/for/mem.raw (synthetic Windows dump), list processes and identify the one that is masquerading (wrong parent/path) as a system process. Report its PID, name, and the anomaly to answer.json.

Platforms
win · Windows lin · Linux
Applications
Volatility
Interfaces
term · Terminal desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.json PID and name match the lab's injected/masquerading process and the anomaly matches the key (e.g. unexpected parent or path). Graded on the reported fields.

Proposed partial credit

40% correct PID; 30% correct name; 30% correct anomaly described

R03.05.01.04-T07 Recover the malicious command line and parent

Original instruction

From fixtures/for/mem.raw, extract the full command line of the suspicious process found earlier and its parent process chain up to the shell/explorer root. Write the chain to chain.json in order.

Platforms
win · Windows lin · Linux
Applications
Volatility
Interfaces
term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

chain.json command line matches the lab value exactly and the parent chain order matches ground truth. Graded on the reported chain.

Proposed partial credit

50% exact command line; 50% correct ordered parent chain

R03.05.01.04-T08 Dump a process and extract its config from memory

Original instruction

From fixtures/for/mem.raw, dump the suspicious process's memory, carve the in-memory configuration blob, decode it, and write the recovered C2 settings to mem_config.json. Note the virtual address range you dumped.

Platforms
win · Windows lin · Linux
Applications
Volatility · Ghidra
Interfaces
term · Terminal desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

mem_config.json field values match the lab ground truth; the noted address range overlaps the config region; dump hash reproducible. Graded on reported config + range.

Proposed partial credit

30% process dumped from correct range; 30% config blob carved; 40% decoded settings correct

R03.05.01.04-T09 Identify the exfil session and export objects

Original instruction

In Wireshark, open fixtures/for/exfil.pcap and find the HTTP session that uploads a file to an external host. Report the client IP, server host, and uploaded filename, and export the transferred object. Save findings to answer.json.

Platforms
win · Windows lin · Linux mac · macOS
Applications
Wireshark
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.json client IP, server host, and filename match the lab key; the exported object's sha256 matches the expected file. tshark cross-check allowed; graded on reported fields + exported object hash.

Proposed partial credit

40% correct session identified; 30% client/host/filename; 30% object exported (hash matches)

R03.05.01.04-T10 Reconstruct a transferred file from a capture

Original instruction

fixtures/for/transfer.pcap contains a file sent over FTP-DATA split across many segments. Reassemble the stream, reconstruct the original file, and report its name and sha256 to answer.json. Save the reconstructed file to out/.

Platforms
win · Windows lin · Linux mac · macOS
Applications
Wireshark
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product
WP09 · Files & folders
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

The reconstructed file in out/ has the sha256 recorded in answer.json and matches the lab's expected file; filename correct. Follow-stream GUI or tshark export both acceptable; graded on file hash.

Proposed partial credit

50% stream reassembled; 30% correct file reconstructed (hash); 20% correct name

R03.05.01.04-T11 Find a file hidden by renamed extension

Original instruction

On fixtures/for/usb.E01, one file has a mismatched extension (header says it is an archive, name says .txt). Use Autopsy's file-type vs extension mismatch view to find it, report its path, and tag it.

Platforms
win · Windows lin · Linux
Applications
Autopsy
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Reported path matches the lab's planted mismatch file and the file is tagged in the case. Graded on the case tag + reported path.

Proposed partial credit

60% correct file identified; 40% tagged in case

R03.05.01.04-T12 Recover persistence from a registry hive

Original instruction

From the exported hives in fixtures/for/hives/, recover autostart persistence: the Run-key entries and the most recently connected USB device. Report entries and the USB serial/first-connect time to answer.json.

Platforms
win · Windows
Applications
Autopsy
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment code · Code & commands
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

answer.json Run-key entries match the lab key and the USB device serial + first-connect time match ground truth. RegRipper CLI or a GUI hive viewer both acceptable; graded on reported fields.

Proposed partial credit

40% Run-key persistence entries; 40% USB device serial; 20% first-connect time

R03.05.01.04-T13 Recover command history from memory

Original instruction

From fixtures/for/mem.raw, recover the console command history for the attacker session and the clipboard contents at capture time. Write the ordered command list to cmds.txt and clipboard text to clip.txt.

Platforms
win · Windows lin · Linux
Applications
Volatility
Interfaces
term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

cmds.txt matches the lab's recorded command history in order and clip.txt matches the captured clipboard. Graded on the two files.

Proposed partial credit

60% command history recovered+ordered; 40% clipboard recovered

R03.05.01.04-T14 Correlate disk and memory evidence

Original instruction

Given fixtures/for/host.E01 and fixtures/for/mem.raw from the same host, link the suspicious process in memory to its on-disk executable and its creation event. Produce correlate.json with the PID, on-disk path+hash, and the disk creation timestamp.

Platforms
win · Windows lin · Linux
Applications
Volatility · Autopsy
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment code · Code & commands
Work product
WP12 · Answer/decision
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

correlate.json PID, on-disk path, file hash, and creation timestamp all match lab ground truth and the hash matches the dumped image. Graded on reported fields across both sources.

Proposed partial credit

30% memory process identified; 30% on-disk binary located (hash); 20% creation time; 20% consistent linkage

R03.05.01.04-T15 Recover a hiding algorithm and decode config

Original instruction

fixtures/for/packed.bin hides its config behind a custom multi-byte XOR+rotate scheme. In Ghidra, recover the exact transform statically, implement decode.py, decode the config, and write it to decoded_config.json. Annotate the transform in Ghidra.

Platforms
win · Windows lin · Linux
Applications
Ghidra
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment code · Code & commands
Work product
WP06 · Code/engineering
Conditions
long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

decode.py reproduces the lab's reference decode on a sample; decoded_config.json matches ground truth; the transform function is labeled in Ghidra. Static only.

Proposed partial credit

40% transform recovered; 30% decode.py correct on sample; 30% config decoded

R03.05.01.04-T16 Extract image metadata indicators

Original instruction

From the carved images in fixtures/for/photos/, extract EXIF metadata and report which images carry GPS coordinates and camera serials. Write indicators.csv (filename, has_gps, lat, lon, camera_serial).

Platforms
win · Windows lin · Linux mac · macOS
Applications
Autopsy
Interfaces
term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

indicators.csv rows match the lab ground truth for which files have GPS and the coordinate/serial values (within tolerance). Graded on the CSV.

Proposed partial credit

40% correct has_gps flags; 40% coordinates; 20% camera serials

R03.05.01.04-T17 Extract email artifacts from a disk image

Original instruction

fixtures/for/mail.E01 contains a mailbox export. Recover the phishing email that started the incident: report sender, subject, attachment name, and its sha256. Save to email.json and tag the artifact in the case.

Platforms
win · Windows
Applications
Autopsy
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

email.json sender/subject/attachment name match the lab key and the attachment hash matches; artifact tagged in case. Graded on reported fields + tag.

Proposed partial credit

30% email located+tagged; 40% sender/subject/attachment; 30% attachment hash

R03.05.01.04-T18 List dump processes via Volatility CLI (control pair of T06)

Original instruction

Control pair of T06: using only the Volatility command line (no GUI), run pslist/psscan over fixtures/for/mem.raw and write the full process table (PID, PPID, name, create time) to pslist.csv. Flag the row that psscan reveals but pslist hides.

Platforms
lin · Linux
Applications
Volatility
Interfaces
term · Terminal
GUI need · task setting
CL · CLI-sufficient
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

pslist.csv matches the lab's process table and the flagged hidden/unlinked process matches the one psscan-only reveals. Graded on the CSV; pure-CLI route, contrast with the T06 GUI/hybrid approach.

Proposed partial credit

50% complete process table; 30% correct fields; 20% hidden process flagged

R03.05.01.04-T19 Prove execution from Prefetch evidence

Original instruction

From fixtures/for/win.E01, parse Windows Prefetch to prove the suspicious binary ran: report its run count, last run time, and the volume it ran from. Write to exec.json and build a short execution timeline note.

Platforms
win · Windows
Applications
Autopsy
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment code · Code & commands
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

exec.json run count, last run time, and source volume match the lab ground truth. Prefetch GUI viewer or CLI parser acceptable; graded on reported fields.

Proposed partial credit

40% correct binary's prefetch found; 30% run count+last run; 30% source volume + note

R03.05.01.04-T20 Produce a forensic IOC report from image and dump

Original instruction

Given fixtures/for/host.E01 and fixtures/for/mem.raw, investigate the intrusion end to end and deliver iocs.json (per fixtures/for/ioc_schema.json: hashes, C2, persistence, accounts) plus report.md with a timeline and findings. Tag all supporting artifacts in the Autopsy case.

Platforms
win · Windows lin · Linux
Applications
Autopsy · Volatility · Ghidra
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
HY · Hybrid
Interaction primitives
nav · Pan/zoom/scroll vis · Visual judgment code · Code & commands
Work product
WP01 · Document
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

iocs.json validates against the schema and its entries match the lab ground truth (>=80% of fields); report.md timeline ordering is correct (rubric for narrative); case contains the supporting tags. Graded on the IOC JSON + tags + rubric.

Proposed partial credit

40% IOC fields correct; 30% correct timeline; 30% artifacts tagged + report sections present

Mapped research references

Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.

In this workflow family