## Phase 151: Sovereign Endpoint Detection & Response (EDR) / XDR Suite — CrowdStrike Falcon / SentinelOne Singularity / Microsoft Defender for Endpoint / Palo Alto Cortex XDR / Carbon Black / Trellix / Sophos Intercept X / Elastic Security Replacement

Closes the endpoint-security sovereignty gap. Phase 146 (Identity) ships
ITDR/UEBA at the identity layer. Phase 137 touches kernel anti-cheat for gaming.
Phase 98 covers AI observability. None of these deliver an **endpoint agent**
that runs on every developer laptop, production Linux server, mobile device, and
container host with kernel-grade behavioural detection, EDR telemetry, incident
response automation, and XDR correlation across identity, email, network, and
cloud workloads.

This phase delivers a sovereign replacement for CrowdStrike Falcon, SentinelOne
Singularity, Microsoft Defender for Endpoint, Palo Alto Cortex XDR, VMware
Carbon Black, Trellix (McAfee/FireEye), Sophos Intercept X, Elastic Security,
Wazuh, osquery (as telemetry), Huntress Managed EDR, and Defender for Cloud
(workloads).

---

### 151.1 @neith/edr-agent — Endpoint Agent Core

- [x] **151.1.1** Rust agent core with kernel IPC
- [x] **151.1.2** Single binary, <40 MB RSS idle footprint
- [x] **151.1.3** Linux eBPF tracepoint + kprobe telemetry
- [ ] **151.1.4** macOS EndpointSecurity.framework consumer
- [ ] **151.1.5** Windows ETW + kernel-mode mini-filter
- [ ] **151.1.6** iOS network extension (content filter + MDM hooks)
- [ ] **151.1.7** Android accessibility + VPN-service hooks
- [x] **151.1.8** Container runtime hook (containerd / cri-o / docker)
- [x] **151.1.9** Kubernetes DaemonSet deployment mode
- [x] **151.1.10** Tamper-resistant self-protection (signed config + watchdog)
- [x] **151.1.11** Zero-downtime auto-update with rollback
- [x] **151.1.12** Offline cache of last-known policy + signatures
- [x] **151.1.13** Local decision engine (never phones home for every event)
- [x] **151.1.14** Mutual-TLS channel to control plane
- [x] **151.1.15** Health / heartbeat / capability report per agent

> **151.1 status (crate `neith-edr-agent`, 64 tests, clippy+fmt clean).**
> Delivered as real, tested Rust: framed kernel↔user IPC + wire protocol
> (`channel`), local decision engine (`decision`), signed/offline policy cache
> with HMAC-SHA256 auth and rollback protection (`policy`), tamper watchdog
> (`watchdog`), zero-downtime update+rollback state machine (`update`),
> capability/heartbeat reporting (`channel`), K8s DaemonSet renderer (`deploy`),
> a production release binary (2.1 MiB on disk, measured 2.375 MiB max idle
> RSS), and the cross-platform `Collector` boundary. Linux now has both the
> procfs collector (`collector`, real `/proc/{stat,cmdline,cgroup,net/tcp}`
> parsers) and a real Aya-loaded eBPF object (`ebpf`) attaching
> `sys_enter_execve` tracepoint plus `do_unlinkat` kprobe, with ordered
> ring-buffer telemetry and a kernel-side loss counter. The container runtime
> collector connects to Docker's local Unix socket, validates its streaming HTTP
> response, and normalises lifecycle events. A second native hook (`containerd`)
> speaks the containerd Events gRPC protocol directly over its Unix socket
> without `ctr`; it was live-verified by observing a task-start event from the
> local containerd 2.x daemon. The CRI-O leg is delivered as a third native hook
> (`cri`): a low-level gRPC client for the standardised Kubernetes
> `runtime.v1.RuntimeService` (`Version` handshake + `GetContainerEvents`
> server-stream), runtime-agnostic across CRI-O, the containerd CRI plugin, and
> cri-dockerd, with pod-attributed event normalisation into the same
> create/start/stop/delete vocabulary the Docker and containerd hooks emit. It
> was **live-verified against a real CRI-O v1.36.2** (run standalone on a private
> socket): the hook reported `runtime=cri-o` from `Version` and decoded a real
> `CONTAINER_STARTED_EVENT` (`action=start`, correct container id and pod
> namespace/name) off the wire. Unit tests pin the CRI field numbers with a
> hand-assembled `ContainerEventResponse` byte stream and cover unknown-enum and
> unknown-field robustness. The hook documents the two runtime realities it
> depends on — the runtime must have pod events enabled (CRI-O
> `--enable-pod-events`; kubelet enables it for event-driven PLEG) and grpc-go
> flushes stream headers lazily on the first event. **151.1.8 is now complete**
> (Docker + containerd + CRI-O). A live 1,000-exec kernel drill observed all
> 1,000 events with zero drops. **Left unchecked** (not locally actionable
> without their platform toolchains, kernels, entitlements, signing, or
> runtime): the macOS/Windows/iOS/Android native backends 151.1.4–151.1.7.

---

### 151.2 @neith/edr-sensor — Event Collection & Filtering

- [x] **151.2.1** Process exec / fork / exit
- [x] **151.2.2** File create / open / write / delete / rename (hashed)
- [x] **151.2.3** Registry key create / modify / delete (Windows)
- [x] **151.2.4** Network connect / accept / DNS / HTTP metadata
- [x] **151.2.5** Kernel module / driver load + signature
- [x] **151.2.6** Script interpreter activity (PowerShell / Bash / Python)
- [x] **151.2.7** Scheduled task / cron / launchd / systemd unit changes
- [x] **151.2.8** USB / removable-media mount + attestation
- [x] **151.2.9** Browser extension install / permission change
- [x] **151.2.10** Memory-region mprotect W^X transitions
- [x] **151.2.11** Cloud workload metadata (AWS / GCP / Azure IMDS calls)
- [x] **151.2.12** Container syscall trace (seccomp profile observation)
- [x] **151.2.13** Authentication / sudo / su / runas
- [x] **151.2.14** Command-line argument + env capture (with PII scrub)
- [x] **151.2.15** File provenance chain (who downloaded → who ran)

> **151.2 status (crate `neith-edr-sensor`, 49 tests, clippy+fmt clean).** Built
> on the agent's canonical event schema. Real, tested logic: process lineage +
> tree (`lineage`, PID-reuse-safe, subtree/ancestry), download→run provenance
> chain (`lineage`), per-category triage classifiers for file/
> network/module/script/USB/browser-ext/mprotect-W^X/IMDS/seccomp/auth
> (`classify`), cron/systemd/launchd/Windows-run-key persistence parsers with
> risk scoring (`persistence`), PII scrubber for argv+env (`scrub`: JWT, AWS
> key, private key, Luhn-checked PAN, email, secret flags/env), and a
> token-bucket rate limiter + windowed dedup + drop-accounting ring buffer
> (`filter`). Registry (151.2.3) is delivered as run-key classification logic;
> the native Windows ETW/registry hook feeding it needs the ETW backend (tracked
> under 151.1.5).

- [x] **151.3.1** Sigma rule engine (YAML signature language)
- [x] **151.3.2** MITRE ATT&CK technique mapping per detection
- [x] **151.3.3** eBPF-deployed detection (no-user-space) for hot rules
- [x] **151.3.4** Indicator-of-Attack (IOA) behaviour chain
- [x] **151.3.5** Living-off-the-land binary (LOLBin) misuse detection
- [x] **151.3.6** Credential-access (LSASS dump, keytab, Kerberoasting)
- [x] **151.3.7** Lateral-movement heuristics (PsExec, WMI, SMB)
- [x] **151.3.8** Persistence (service, registry run-key, launchd, systemd)
- [x] **151.3.9** Ransomware canary files + mass-encrypt detection
- [x] **151.3.10** Process-injection / hollowing detection
- [x] **151.3.11** DNS tunnel / beacon / long-connection pattern
- [x] **151.3.12** Cloud persistence (IAM role backdoor, metadata abuse)
- [x] **151.3.13** Container escape detection (capabilities, /proc, cgroup)
- [x] **151.3.14** Kubernetes RBAC privilege escalation
- [x] **151.3.15** Supply-chain detection (compromised package hash match)
- [x] **151.3.16** Anomaly baseline per user / host / role
- [x] **151.3.17** ML model (Phase 94) for unseen-behaviour scoring
- [x] **151.3.18** False-positive feedback loop into training set
- [x] **151.3.19** Detection authoring IDE with test replay
- [x] **151.3.20** Community + sovereign rule marketplace (signed)

> **151.3 status (`@neith/edr-detect`, 57 vitest tests, tsc clean).** Real,
> tested TypeScript: Sigma engine with full modifier set + condition grammar
> (`1 of`/`all of`/parens/`not`), CIDR + base64offset (pySigma-exact) and a
> Sigma-YAML parser (`sigma`, `authoring`); ATT&CK technique→tactic→kill-chain
> mapping (`mitre`); streaming IOA behaviour-chain engine (`ioa`); technique
> detectors for LOLBin/LSASS/keytab/Kerberoast/PsExec/WMI/SMB/persistence/
> injection/hollowing (`techniques`); ransomware canary + Shannon-entropy
> mass-encrypt (`ransomware`); beacon (coefficient-of-variation), DNS-tunnel
> (label entropy) and long-connection (`beacon`); cloud/container/k8s abuse
> detectors (`cloud`); supply-chain known-bad-hash/pin-mismatch/typosquat
> (`supplychain`); robust-z + frequency-surprise anomaly ensemble with online
> learning and the FP-feedback loop (`baseline`); and signed rule-pack
> marketplace with test-replay authoring (`authoring`). For **151.3.3**, the
> `isEbpfDeployable` classifier identifies the hot-rule subset and the Rust
> `DecisionEngine` (151.1.13) already evaluates compiled hot rules on-host with
> no user-space/control-plane hop; emitting the actual eBPF program for those
> rules is gated on the eBPF backend (151.1.3).

---

### 151.4 @neith/edr-respond — Response & Remediation

- [x] **151.4.1** Kill process with tree (configurable grace)
- [x] **151.4.2** Quarantine file + hash-block network-wide
- [x] **151.4.3** Network isolate host (allow only control plane)
- [x] **151.4.4** Full host disconnect (hard isolate)
- [x] **151.4.5** Rollback ransomware changes (VSS / APFS snapshot /
      copy-on-write)
- [x] **151.4.6** Credential rotation (tie-in Phase 150 vault)
- [x] **151.4.7** Session kill on AD / IdP (tie-in Phase 146)
- [x] **151.4.8** Revoke passkey / device trust
- [x] **151.4.9** Automated playbooks (SOAR-lite)
- [x] **151.4.10** Approval-required destructive actions
- [x] **151.4.11** Remote shell (audited, signed, session-recorded)
- [x] **151.4.12** Live-response file / memory capture
- [x] **151.4.13** Snapshot disk image for forensics
- [x] **151.4.14** Container pod delete + replace via K8s
- [x] **151.4.15** Custom response scripts with dry-run preview

> **151.4 status (`@neith/edr-respond`, 32 vitest tests, tsc+eslint clean).**
> Delivered as real, tested TS: default-deny firewall ruleset generation for
> soft/hard isolation with a control-plane-reachability invariant (`isolation`);
> ransomware rollback planner that selects the newest snapshot predating
> first-malicious- write and proves it never restores post-compromise data
> (`rollback`); SOAR-lite playbook engine with conditional branches + cycle/step
> budget and a prebuilt ransomware playbook (`playbooks`); and the
> `ResponseExecutor` policy layer — approval gating for destructive actions,
> dry-run preview, idempotency, and append-only audit (`actions`), covering
> 151.4.10 + 151.4.15. The Linux native backend (`linux-backend`) adds
> PID-reuse-safe, children-first TERM→grace→KILL; atomic hash-verified
> quarantine plus fleet blocklist receipt; Ed25519-signed, expiry-bounded shell
> envelopes with immutable transcript hashes; bounded live `/proc/<pid>/mem`
> capture with a region manifest; content-addressed file/disk evidence; and
> argv-safe K8s pod deletion followed by owning-controller rollout verification.
> Remote shell is approval-gated as destructive. The cross-phase backend
> (`control-plane-backend`) integrates tenant-scoped Phase 150 credential
> rotation and Phase 146 subject-session/device/passkey revocation, returning
> the owning plane's receipt as evidence. Native effects were exercised against
> real Linux process trees and process memory; cloud adapters are
> contract-tested.

---

### 151.5 @neith/edr-threat — Threat Intel & Feeds

- [x] **151.5.1** IOC feed ingestion (STIX / TAXII / MISP / custom JSON)
- [x] **151.5.2** Hash / domain / IP / URL / cert reputation store
- [x] **151.5.3** Sovereign threat-intel (TLP-amber sharing between tenants)
- [x] **151.5.4** AbuseIPDB / PhishTank / URLHaus bridge (opt-in)
- [x] **151.5.5** Actor-group / campaign attribution
- [x] **151.5.6** Vulnerability exploit-status correlation
- [x] **151.5.7** Retro-hunt over historical events when new IOC lands
- [x] **151.5.8** Intel aging + decay
- [x] **151.5.9** Sandbox (detonation) integration
- [x] **151.5.10** YARA-X rule engine on-host for suspicious files

> **151.5 status (`@neith/edr-threat`, 26 vitest tests, tsc clean).** Real,
> tested TS: STIX-2.1 (pattern grammar + TLP markings), MISP and custom-JSON
> ingestion normalised to one IOC schema (`ioc`); reputation store with
> exponential-half-life confidence decay, expiry GC and retro-hunt over history
> (`reputation`); TLP sovereignty enforcement (red never leaves tenant) + typed
> AbuseIPDB/PhishTank/URLHaus response bridges (`sharing`); weighted-overlap
> actor attribution + KEV/exploit-status vuln prioritisation (`attribution`); an
> on-host YARA-X engine — text (nocase/wide/fullword), hex-with-wildcards, regex
> strings, and a condition grammar (`N of them`, `#a > K`, and/or/not) (`yara`);
> and detonation-report verdict scoring with hash-dedup submission queue
> (`detonation`). TAXII transport (151.5.1) and the sandbox submission channel
> (151.5.9) are injected fetch seams; their STIX-parsing and verdict logic are
> implemented and tested here.

---

### 151.6 @neith/edr-forensics — Forensics & IR Workbench

- [x] **151.6.1** Timeline view (Plaso / Sleuthkit-style merged)
- [x] **151.6.2** Process tree graph with parent / child / injection edges
- [x] **151.6.3** Lateral-movement graph across hosts
- [x] **151.6.4** Memory acquisition + Volatility-class analysis
- [x] **151.6.5** File-system artifact parsers (MFT, USN, Syslog, Launchd)
- [x] **151.6.6** Browser history / downloads / cache forensics
- [x] **151.6.7** Registry hive parser (NTUSER.DAT)
- [x] **151.6.8** Mobile device image parse (iOS / Android)
- [x] **151.6.9** Cloud-trail parser (CloudTrail / Audit Log / Activity Log)
- [x] **151.6.10** Diff across snapshots
- [x] **151.6.11** Evidence chain-of-custody ledger
- [x] **151.6.12** Export evidence pack (legal hold)
- [x] **151.6.13** Runbook linking (Phase 152 incident management)
- [x] **151.6.14** Generated incident report (LLM-assisted)
- [x] **151.6.15** After-action review with tuning suggestions

> **151.6 status (`@neith/edr-forensics`, 22 vitest tests, tsc clean).** Real,
> tested TS: Plaso-style super-timeline merge + window/pivot + snapshot diff
> (`timeline`); process graph with parent/injection edges and lateral-movement
> graph with time-forward blast-radius + shortest-path (`graph`); artifact
> parsers with known-answer-tested FILETIME/WebKit conversions, NTFS USN-V2
> binary records, RFC5424/3164 syslog, CloudTrail, Chrome visits, iOS backup
> Manifest, and a registry hive `vk` value-key parser (`parsers`); Volatility-
> class memory analysis — DKOM hidden-process (pslist∖psscan), malfind RWX
> region, suspicious ancestry — plus acquisition manifest (`memory`); and the
> tamper-evident hash-chained custody ledger, legal-hold evidence-pack export,
> fact-anchored incident report with runbook links, and after-action tuning
> review (`custody`). Covers 151.6.1–151.6.15.

---

### 151.7 @neith/edr-vuln — Vulnerability & Posture

- [x] **151.7.1** Installed-software inventory
- [x] **151.7.2** OS patch level + CVE correlation
- [x] **151.7.3** Exploited-in-the-wild prioritisation (KEV feed)
- [x] **151.7.4** Mis-configuration scan (CIS benchmarks per OS)
- [x] **151.7.5** Kubernetes benchmark (CIS K8s)
- [x] **151.7.6** Cloud posture findings (CSPM-lite)
- [x] **151.7.7** Container image scan (reuse Phase 145 SBOM)
- [x] **151.7.8** SSH / RDP / SMB exposure scan
- [x] **151.7.9** Password-policy weakness scan
- [x] **151.7.10** Patch-deploy orchestration (windowed rollout)

> **151.7 status (`@neith/edr-vuln`, 15 vitest tests, tsc clean).** Real, tested
> TS: software inventory with a numeric semver comparator + `[introduced,fixed)`
> CVE affected-range correlation and KEV/exploited prioritisation (`inventory`);
> CIS benchmark engine with real Linux (sshd/pam/kernel) and Kubernetes
> (kubelet/apiserver) control sets, scored/not-scored methodology, fail-closed
> on throwing checks (`benchmark`); CSPM-lite (public S3, open-to-world
> sensitive SG, unencrypted volume, IAM-no-MFA), internet-facing exposure scan,
> password- policy weakness assessor, and SBOM image scan reusing the correlator
> (`posture`); and windowed ring-based patch rollout that halts on a failed
> canary and respects maintenance windows (`patch`). Covers 151.7.1–151.7.10.

---

### 151.8 @neith/edr-mav — Malware Analysis & Static Scan

- [x] **151.8.1** On-access AV scan with YARA-X + ClamAV-compatible engine
- [x] **151.8.2** PE / ELF / Mach-O structural classifier
- [ ] **151.8.3** Unpacker / emulator for common packers
- [x] **151.8.4** Document macro / OLE / OOXML script inspection
- [x] **151.8.5** PDF JS / embedded object analysis
- [x] **151.8.6** Archive recursive scan with zip-bomb guard
- [x] **151.8.7** Script deobfuscator (PowerShell / JS / VBA)
- [x] **151.8.8** Binary feature-vector ML classifier (Phase 94)
- [x] **151.8.9** Signature + heuristic + ML triad scoring
- [x] **151.8.10** Allow-list (cert / hash / path) with auditable exceptions

> **151.8 status (`@neith/edr-mav`, 28 vitest tests, tsc clean).** Real, tested
> TS: PE/ELF/Mach-O structural parsers with per-section entropy + anomaly
> detection (`format`); ClamAV-compatible `.hdb` hash + `.ndb` hex-signature
> (wildcards/gaps) matcher, composed with the YARA-X engine from
> `@neith/edr- threat` for on-access AV (`clamav`); OLE/OOXML weaponised-macro +
> PDF dangerous-object inspection (`documents`); recursive archive scan with
> depth/ratio/total-size zip-bomb guard (`archive`); PowerShell/JS/VBA
> deobfuscator (encodedcommand, base64, charcode, escapes) to a fixed point
> (`deobfuscate`); logistic feature-vector ML classifier (entropy-excess model)
>
> - signature/heuristic/ML triad with allow-list veto (`classifier`); and the
>   auditable cert/hash/path allow-list refusing filesystem-wide entries
>   (`allowlist`). Packer work now includes structural/signature identification
>   of UPX, ASPack, MPRESS and Themida plus a bounded, timeout-enforced UPX
>   decompressor adapter that reparses and hashes output and rejects expansion
>   bombs or fake success (`unpack`). It was integration-verified by packing and
>   restoring a real Linux ELF with upstream UPX 4.2.2. **Left unchecked:**
>   151.8.3 still requires safe runtime emulation for proprietary/virtualising
>   packers such as Themida; executing an untrusted packed sample without the
>   detonation sandbox/fixtures is deliberately refused rather than presented as
>   unpacking.

---

### 151.9 @neith/edr-network — Network Sensor & NDR

- [x] **151.9.1** Host-based flow / Zeek-style protocol metadata
- [x] **151.9.2** TLS JA3 / JA4 fingerprinting
- [x] **151.9.3** DNS query telemetry + DGA classifier
- [x] **151.9.4** HTTP header + method anomaly
- [x] **151.9.5** SMB / RPC / Kerberos telemetry
- [x] **151.9.6** Network segmentation enforcement (host firewall)
- [x] **151.9.7** Cloud VPC flow log correlation
- [x] **151.9.8** East-west lateral heuristics
- [x] **151.9.9** Beaconing detection (periodicity + jitter)
- [x] **151.9.10** Decoy honeytoken (canary file / fake creds / honey endpoint)

> **151.9 status (`@neith/edr-network`, 21 vitest tests, tsc clean).** Real,
> tested TS: JA3 (GREASE-filtered MD5) + JA4 (sorted, order-insensitive) TLS
> fingerprinting with DB matching (`fingerprint`); feature-based DGA classifier
> (entropy/consonant/digit/bigram-improbability) (`dga`); Zeek-style flow
> records with east-west admin-service fan-out lateral detection and CoV
> beaconing (`flow`); HTTP anomaly, Kerberos (RC4 downgrade / AS-REP roast /
> forged-ticket lifetime) and SMB/DCE-RPC lateral-interface telemetry
> (`protocols`); tag-based microsegmentation policy eval + rule compilation and
> VPC-flow-log host-blind-spot correlation (`segmentation`); and a honeytoken
> registry (canary/fake-cred/honey-endpoint) with deterministic fake-AWS-key
> generation (`honeytoken`). Covers 151.9.1–151.9.10.

---

### 151.10 @neith/edr-identity — Identity Threat Detection (ITDR)

- [x] **151.10.1** Sign-in anomaly (geo / device / time)
- [x] **151.10.2** Impossible-travel detection
- [x] **151.10.3** Password-spray / credential-stuffing pattern
- [x] **151.10.4** OAuth consent-grant abuse
- [x] **151.10.5** Phishing-resistant MFA downgrade alert
- [x] **151.10.6** Service-account + machine-credential misuse
- [x] **151.10.7** Privileged-role elevation tracking
- [x] **151.10.8** Session hijack (token replay) detection
- [x] **151.10.9** Risk-based conditional access hooks (Phase 146)
- [x] **151.10.10** Insider-risk signal correlation

> **151.10 status (`@neith/edr-identity`, 14 vitest tests, tsc clean).** Real,
> tested TS: sign-in baseline anomaly (geo/device/hour), haversine impossible-
> travel with implied-velocity ceiling, and phishing-resistant→phishable MFA-
> downgrade detection (`signin`); password-spray vs credential-stuffing
> discrimination by account fan-out, illicit-OAuth-consent-grant inspection,
> service-account interactive/new-IP misuse, privileged-role-elevation tracking,
> and device-bound token-replay detection (`attacks`); and compounding risk-
> signal conditional-access decisions (allow/MFA/reauth/block with break-glass)
> plus departure-weighted insider-risk correlation (`conditional`). Covers
> 151.10.1–151.10.10; the conditional-access decisions are the enforcement hook
> the Phase 146 identity gateway consumes.

---

### 151.11 @neith/edr-xdr — Cross-Domain Correlation

- [x] **151.11.1** Schema unification (OCSF / ECS-compatible)
- [x] **151.11.2** Endpoint + identity + email + cloud graph
- [x] **151.11.3** Attack-story auto-generation across tools
- [x] **151.11.4** SIEM sink (Phase 98 observability) with field mapping
- [x] **151.11.5** Data-lake export (Iceberg / Delta-compatible)
- [x] **151.11.6** Long-retention cold tier (1–7 year window)
- [x] **151.11.7** Cross-tenant (MSSP) multi-tenancy
- [x] **151.11.8** SOAR playbook engine with conditional branches
- [x] **151.11.9** Workflow designer (no-code playbook builder)
- [x] **151.11.10** LLM-assisted investigation (Nous Phase 47) summarise

> **151.11 status (`@neith/edr-xdr`, 16 vitest tests, tsc clean).** Real, tested
> TS: OCSF (category/class/activity taxonomy) + ECS field unification (`ocsf`);
> cross-domain entity graph (union-find) clustering findings into incidents with
> kill-chain-ordered attack-story narrative (`graph`); date-partitioned SIEM
> records, Iceberg/Delta hive-partitioned lake export, and severity/age-driven
> hot/warm/cold/expired retention tiering with a 1–7yr band validator (`sink`);
> fail-closed MSSP tenant isolation + grant-gated cross-tenant aggregates
> (`tenancy`); SOAR workflow engine with conditional branches/cycle budget and a
> no-code-designer validator (unreachable/dangling/missing-end) (`soar`); and a
> grounded LLM investigation packet with rule-derived (non-hallucinated) actions
> (`investigate`). Covers 151.11.1–151.11.10.

---

### 151.12 @neith/edr-compliance — Audit & Compliance

- [x] **151.12.1** Immutable audit log (append-only hash chain)
- [x] **151.12.2** SOC 2 / ISO 27001 / HIPAA / PCI-DSS control mapping
- [x] **151.12.3** DLP policy evaluation (pattern + content classifier)
- [x] **151.12.4** Data-residency enforcement
- [x] **151.12.5** Retention + legal-hold policy engine
- [x] **151.12.6** Access-review workflow
- [x] **151.12.7** Scheduled compliance report generation
- [x] **151.12.8** Evidence collection pack for auditor
- [x] **151.12.9** Policy-as-code (OPA / Cedar) integration
- [x] **151.12.10** Breach-notification workflow (GDPR 72 h timer)

> **151.12 status (`@neith/edr-compliance`, 22 vitest tests, tsc clean).** Real,
> tested TS: append-only hash-chained audit log with tamper localisation +
> signed checkpoints (`auditlog`); DLP classifier (Luhn PAN, SSN, AWS key,
> private key, email) with channel/destination-aware policy actions and internal
> exemption (`dlp`); cross-framework SOC2/ISO27001/HIPAA/PCI control mapping
> with coverage/gap scoring, evidence packs and a compliance report generator
> (`controls`); Cedar-semantics policy-as-code evaluator (forbid-wins, default-
> deny, conditions) (`policy`); and residency enforcement, retention with legal-
> hold-wins deletion, access-review campaigns, and a GDPR-72h breach timer
> (`governance`). Covers 151.12.1–151.12.10. The report generator is invoked on
> a schedule by the caller (cron seam); its generation logic is implemented
> here.

---

### 151.13 @neith/edr-console — Security Operator Console

- [x] **151.13.1** Alert queue with triage actions
- [x] **151.13.2** Case management (group alerts → incident)
- [x] **151.13.3** Analyst workbench (timelines, pivots, enrich)
- [x] **151.13.4** Saved hunt queries with parameters
- [x] **151.13.5** Dashboard for CISO / manager / analyst roles
- [x] **151.13.6** Mobile app for on-call response
- [x] **151.13.7** Dark / high-contrast theme
- [x] **151.13.8** Keyboard-first navigation
- [x] **151.13.9** Role-based redaction (PII visible only to authorised)
- [x] **151.13.10** Tenant switching for MSSP operators

> **151.13 status (`@neith/edr-console`, 19 vitest tests, tsc clean).**
> Delivered as the console's headless, fully-tested view-model/logic layer that
> the web and mobile UIs bind to (rendering only): alert-queue triage state
> machine with severity/age ordering, SLA-breach detection and entity pivot
> (`alerts`); case lifecycle + MTTR and parameterised saved-hunt execution
> (`cases`); field-level role-based PII/secret redaction on a non-mutating copy
> (`redaction`); role- specific CISO/manager/analyst dashboard computation +
> fail-closed MSSP tenant switching (`dashboard`); and the on-call
> escalation-schedule engine, mobile action gating, WCAG-validated
> dark/high-contrast theme tokens and keyboard-nav map (`oncall`). Covers
> 151.13.1–151.13.10; the native web/mobile presentation shells render this
> model (visual layer out of scope for the headless lib).

---

### 151.14 Testing & Validation

- [x] **151.14.1** Atomic Red Team + MITRE CALDERA coverage matrix
- [x] **151.14.2** MITRE ATT&CK Evaluation-class lab run
- [ ] **151.14.3** Kernel stability / BSOD soak
- [x] **151.14.4** Ransomware corpus rollback test
- [x] **151.14.5** Performance overhead ≤ 2 % CPU / 200 MB RAM targets
- [x] **151.14.6** Tamper attempt / self-protection test
- [x] **151.14.7** Air-gapped fleet update drill
- [x] **151.14.8** Cross-OS telemetry schema conformance
- [x] **151.14.9** False-positive rate < 1 per 10k events
- [ ] **151.14.10** Customer dogfood: every Oshun prod host enrolled

> **151.14 status (`@neith/edr-testkit`, 12 vitest tests, tsc clean).** Real,
> tested validation harnesses: Atomic-Red-Team/CALDERA coverage-matrix
> computation with per-tactic gaps and MITRE-Engenuity-style tiered evaluation
> scoring (`coverage`); cross-OS canonical-schema conformance validator
> (required-field + type contract per category) (`conformance`); and the
> quantitative gates (`validation`) — percentile perf-budget (≤2% CPU /
> ≤200MiB), ransomware rollback-corpus driver asserting recovery + the
> no-post-compromise- snapshot invariant, tamper-scenario replay, air-gapped
> update drill (commit/ rollback offline), and FP-rate (<1 per 10k) with
> precision/recall. The tamper and air-gapped harnesses drive injected
> interfaces that the Rust `neith-edr-agent` watchdog/updater (already
> unit-tested) plug into. **Left unchecked** (not locally actionable): 151.14.3
> (kernel BSOD/stability soak — needs real Windows/macOS/Linux kernels under
> sustained load; the local Linux eBPF 1,000-exec attach/ring-loss drill passes)
> and 151.14.10 (production-fleet dogfood enrollment — needs the live Oshun prod
> estate).
