Wire the policy, feeds, containment, exec guard, behavior, governor, crash safety and event store into the daemon, detection pipeline, CLI and meson build (zstd now required); Dockerfile, conan, systemd unit (LimitCORE=0, resource limits) and installer updates; existing tests adapted (store paths are directories now). Detections are counted and notified even when evidence cannot be persisted (persist_failures in status). prevention_timeout_ms is clamped and behavior dedupe handles out-of-order events. Docs: API contract, developer guide, plan, UX, agent rules. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
4 KiB
Analyst UX Notes
Working notes for Strawberry's operator-facing search and investigation surface.
Linked Documents
Current State
As of October 24, 2026:
- Strawberry's operator surface is a Qt6 desktop app with a tray indicator
(
strawberry-tray). It replaced the temporary Python web evaluator, which was removed together with its HTTP port. - The tray shows scanner health (gray unreachable, green healthy, yellow degraded or unseen low/medium finding, red unseen high/critical finding). It raises desktop notifications for new high/critical findings; clicking one opens the finding.
- Window tabs:
- Findings: severity filter, unseen-only, mark seen, details, pivots to processes
- Processes: search, identity, deleted/memfd flags, lineage, related evidence, incident trace
- Diagnostics: backend, tracepoints, rules, counters, plus reload/rescan controls
- "Seen" state is per desktop user (QSettings) and never touches the daemon.
- Strawberry is still process-centric, not thread-centric.
- Finding, process-search and trace output are also available as CLI JSON for automation; see API_CONTRACT.md.
- Strawberry does not yet have a general multi-rule incident correlation engine.
UX Goal
The target workflow should be:
- search for a process, executable image, hash, rule, or finding
- inspect the correlated summary first
- pivot into lineage, related files, and related indicators
- expand into raw edges and nodes only when needed
Direction: Native Desktop App
The operator surface is a native Qt app rather than a web UI. The reasons:
- No listening port. A local web server is reachable by every local user and, through DNS rebinding, by websites the analyst visits. The app talks to the daemon only over a group-restricted Unix socket.
- Least privilege. The app runs as the desktop user, reads evidence
read-only, and can trigger only the fixed control verbs (
status,reload-rules,rescan). - Same process as the data model. The app links the existing C++
Database,ProcessQueryandIncidentReportercode directly, so there is no second API layer to keep in sync. - Ambient awareness. A tray indicator plus notifications fits a scanner that runs in the background better than a page you have to keep open.
Remote or headless hosts are out of scope for now. If needed, they should get a separate, authenticated interface designed for that purpose, not a loopback web server exposed through tunnels.
Suggested v1 Analyst Views
1) Findings
- finding ID
- severity / score
- short explanation
- first seen / last seen
- primary process
- related indicator count
2) Process search
- PID
- executable path
- hash
- command line
- parent PID
- timestamp
3) Process detail
- process metadata
- parent/child lineage
- executed file image
- matched indicators
- related correlated findings
4) Indicator detail
- rule name
- hash
- source / tags / metadata
- linked process executions
- linked executable images
Thread Support
“Find threads” is a valid long-term UX goal, but it should be treated as post-v1 unless Strawberry first gains:
- TID/thread telemetry
- thread-to-process relationships
- per-thread execution or activity context
Until then, the safest UX is:
- process-first
- lineage-aware
- evidence-pivot friendly
Correlation Engine Relationship
The UX should not sit directly on raw nodes and edges alone.
It should primarily consume:
- correlated findings
- process-centric pivots
- explainable summaries
Findings, process search, lineage, traces and diagnostics now exist in the desktop app. The next UX dependencies are richer evidence (argv, file-write and network telemetry) and an agent-facing interface over the same read-only evidence and control verbs.
Timestamp note: event times are wall-clock nanoseconds (eBPF timestamps are converted from boot time); synthetic demo data may still use small raw values.