strawberry/docs/false-positives.md
trvon fb6b905860
Some checks failed
Build and Test / build (debug, clang) (push) Has been cancelled
Build and Test / build (debug, gcc) (push) Has been cancelled
Build and Test / build (release, clang) (push) Has been cancelled
Build and Test / build (release, gcc) (push) Has been cancelled
Build and Test / docker (push) Has been cancelled
CLI: yams-style commands, with doctor and doctor fix
strawberry [--config PATH] [--json] <command> [<subcommand>] [args] [--flags]

status, doctor, doctor fix <action>, daemon run, findings list/show/
correlate, process search/trace/trace-node/suspend/resume/kill, scan
file/processes/memory, rules reload/update/check/fetch, policy show/set,
opensnitch enable-checksums, store verify/compact, replay, demo seed,
help [command], version.

One table in src/cli_commands.cpp defines each command (words,
positionals, flags, help) and how it maps onto the existing flag form,
which main() still parses; help is generated from the same table.
Unknown options and wrong arity print the command's usage. Process
actions validate node ids before building their JSON payload.

The older flag form (strawberry --status, strawberry <config>) still
works. The systemd units now run `strawberry daemon run --config ...` and
`strawberry rules fetch --config ...`. README, AGENTS.md and the
false-positive guide use the command form.

doctor renders the daemon's checks with their fixes; doctor fix runs a
maintenance action (with --dry-run). Against an older daemon the CLI says
to reinstall rather than "unknown command". replay explains that
telemetry is root-only instead of reporting zeros, and keeps config
output out of --json.

tests/test_cli.cpp: translations, flag-form passthrough, errors, help.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:32:02 -06:00

70 lines
4.1 KiB
Markdown

# Expected false positives
Triage of every finding on a developer workstation (Arch, 2026-10-25: 26
findings, all low/medium, none malicious). Use it to recognise routine
findings and to decide on allowlist entries. Add new classes here as they
are seen.
Allowlist entries are made in the app (Policy) or with
`strawberry policy set allowlist-add '<json>'`. Prefer the narrowest entry:
a `sha256`, or a `path` scoped to one `rule`.
| Class | Rule (severity) | Typical source | Frequency |
|---|---|---|---|
| Binary replaced while running | `DeletedExecutable` (evidence only since 2026-10-26) | auto-updating apps (Claude Code, browsers, IDEs), package upgrades, reinstalling Strawberry itself | after every update, once per still-running process |
| Strawberry's own install | `PersistenceWrite` (evidence only since 2026-10-26, when the unit matches the shipped one) | `scripts/install.sh` writing `strawberry*.service/.timer/.path` with `/usr/bin/install` | every reinstall (4 per run) |
| Self-debugging / sanitizers | `PtraceAttach` (low since 2026-10-25, was medium) | LeakSanitizer in ASan/fuzz builds, crash handlers attaching to their parent | every sanitizer run that exits |
| Deliberate tests | `PtraceAttach` (medium) | `tests/behavior_live` (attaches to a child on purpose) | live test runs |
| Test marker in memory | `Strawberry_Test_Marker` (low) | the marker string in a buffer of an editor, terminal or chat tool | fixed 2026-10-25 (rule now only matches files that start with it) |
## Details
**Evidence only.** Some indicators have a routine explanation most of the
time. They are stored (status `pipeline.evidence_only` counts them) but make
no finding, alert or containment on their own. Correlation still counts them
when the same process has another signal (YARA/IOC hit, network,
persistence), so they add context where it matters.
**DeletedExecutable.** `/proc/<pid>/exe` ends in `(deleted)` when the file
was replaced on disk after the process started. Updates do this constantly:
11 of 16 hits were Claude Code's auto-update, the rest the Strawberry
binaries after a reinstall. When a new file exists at the same path (an
update), the indicator is evidence only. When nothing is at the path (a
binary that deleted itself after starting) it stays a low finding.
Restarting the updated application clears it.
**PersistenceWrite by the installer.** Root writing systemd units is exactly
what a root persistence implant does, so writes by a generic tool such as
`/usr/bin/install` are reported; package managers, `systemctl` and editors
are exempt. `scripts/install.sh` records the SHA-256 of every unit it
installs in `/var/lib/strawberry/installed-files.sha256` (root-owned; the
daemon ignores the manifest otherwise) before installing them. A write whose
file then matches is evidence only. Any other content at those paths, or a
writer running from a suspicious location, is still reported. The daemon
running during an install is the previous version, so this takes effect from
the install after the one that adds it.
**PtraceAttach.** Debuggers (`gdb`, `lldb`, `strace`, `rr`, `perf`,
`valgrind`, `bpftrace`, `systemd-coredump`, ...) are exempt. A parent and
child running the same binary tracing each other (LeakSanitizer's
StopTheWorld) is reported as low. On a workstation that runs sanitizer or
test builds, a path entry for the build tree scoped to this rule keeps it
quiet: `{"kind":"path","value":"/home/<you>/src/","rule":"PtraceAttach"}`
(trailing `/` = prefix).
## Fixed on 2026-10-25
- Behavior detections ignored `path` and `rule` allowlist entries below high
severity (they only checked `sha256`, which is computed for high only).
- Finding summaries did not say what happened. They now include the
indicator details, e.g. the unit file written or the ptrace target
(`metadata.details`).
- `Strawberry_Test_Marker` matched any process holding the text in memory;
it now needs the marker in the first 256 bytes of a file
(EICAR-style end-to-end check).
- Same-binary parent/child ptrace is low instead of medium.
## Changed on 2026-10-26
- Binaries replaced by an update and Strawberry's own installed units are
evidence only (see above).