SOC-Agent-Traces-10K

Multi-step SOC investigation agent traces in session-trace format. Each record is a complete investigation session: an alert arrives, an analyst agent gathers evidence through nine read-only tools, and closes with a structured JSON triage report.

Instead of single-turn alert → answer pairs, every record captures the full reasoning trajectory:

alert → get_surrounding_events → get_process_tree → lookup_attack
      → search_sigma → get_asset_context → get_related_alerts
      → calculate_risk → retrieve_playbook → triage report

Use cases

• Agent SFT / tool-use training (chat-format traces with structured tool calls)
• Agent evaluation and replay benchmarks
• Triage-decision and confidence-calibration research
• Detection-engineering training data (ATT&CK + Sigma grounded)

Schema

Field	Type	Description
scenario_id	string	Unique scenario identifier
alert	JSON string	Alert object (id, title, severity, source, host, user, description)
environment	JSON string	Organization context (industry, security stack, asset focus)
ground_truth	JSON string	Verdict, ATT&CK techniques, root cause, IOCs
trace	JSON string	Full session as OpenAI-style messages with tool_calls
tool_calls	JSON string	Flattened list of tool invocations with argument/result summaries
evidence	JSON string	Decisive evidence items cited in the report
decision	string	escalate / monitor / close
confidence	float	Analyst confidence 0-1
success	bool	Whether the trace reaches the correct verdict
archetype	string	Scenario archetype (32 families)
verdict	string	malicious / benign
n_tool_calls	int	Number of tool invocations
n_messages	int	Number of session messages

Corpus statistics

• Total traces: 10000
• Verdicts: {"malicious": 8161, "benign": 1839}
• Decisions: {"escalate": 7033, "monitor": 1505, "close": 1462}
• Successful traces: {"True": 8498, "False": 1502}
• Avg tool calls / trace: 6.51 · avg messages / trace: 16.01

Archetype distribution

Archetype	Traces
rdp_bruteforce	482
smb_lateral	468
ad_discovery	466
ps_download_cradle	460
browser_cred_theft	458
registry_persistence	453
wmi_remote	443
cloud_exfil	443
process_injection	441
phish_macro_powershell	440
log_clear	427
masquerade_binary	427
dns_tunnel	426
c2_beacon	403
cloud_account_anomaly	377
dev_build	377
backup_spike	371
pass_the_hash	361
kerberoast	355
vuln_scanner	309
edr_tamper	274
new_service_change_window	192
scheduled_task_persistence	178
admin_ps_maintenance	176
lsass_dump	149
webshell	142
sccm_deployment	136
av_selftest	132
onedrive_bulk_sync	88
ransomware_staging	88
windows_update	52
helpdesk_reset_flood	6

Tool usage

Tool	Calls
lookup_attack	10065
get_surrounding_events	10000
calculate_risk	10000
get_asset_context	9617
search_sigma	8321
get_process_tree	6057
get_related_alerts	5644
retrieve_playbook	5225
search_cve	142

Tools (all read-only simulations)

get_surrounding_events, get_process_tree, lookup_attack, search_sigma, get_asset_context, get_related_alerts, calculate_risk, search_cve, retrieve_playbook. No tool executes commands against real systems; every result is a deterministic lookup over the scenario's synthetic evidence store.

Trace format

trace follows the OpenAI chat-message convention: assistant messages carry tool_calls (id, function name, JSON arguments) and tool results arrive as role="tool" messages linked by tool_call_id. The final assistant message contains the triage report as a fenced JSON block with keys verdict, decision, confidence, techniques, summary, evidence_ids, recommended_actions.

Splits

Stratified by verdict and archetype: train 90% / validation 5% / test 5%. Traces with success=false (~12%) are flawed investigations (missed evidence step → wrong or weak conclusion) intended for contrastive training and evaluation; filter them out for SFT.

Generation methodology

Traces are synthesized by a reference investigation policy executing against a simulated SOC backend (per-scenario evidence stores over 32 attack and false-positive archetypes), with step jitter, dead-end lookups, and analyst persona variation. Ground truth is known by construction. All organizations, hosts, users, IPs, domains, and hashes are synthetic.

Companion assets

• Dataset (sibling): alirezaaminzadeh/soc-agent-traces-100k
• Fine-tuned agent: alirezaaminzadeh/soc-agent-traces-smollm3-3b
• Baseline classifiers: alirezaaminzadeh/soc-traces-baseline
• Live investigation demo: alirezaaminzadeh/soc-agent-traces

License

Apache 2.0. Knowledge-base records are condensed paraphrases of public MITRE ATT&CK, Sigma HQ, and NVD material.