Files
Maven/internal/auth/policy.go
T
kami 7c7bd8ceeb Ship voice enrolment, and report recognition as blocked (#255)
Maven can now be told who someone is. She cannot yet tell who is speaking,
and this commit is careful to say so rather than pretend otherwise.

What works: profiles are enrolled from several deliberately recorded samples,
listed, and deleted. They live in the existing memory_vectors table under a
"speaker:" id prefix, so there is no migration; what that needed was a wider
interface than memory.Store, hence memory.Catalog with ByPrefix and Delete.
Delete is the load-bearing half — a voiceprint someone asked to be rid of has
to actually go, and a search-only store cannot do that. InMemoryStore.Insert
became an upsert by id to match what the persistent store already did.

What does not work, and why it is not faked: there is no speaker-embedding
model on this box. Sixteen ggufs in /mnt/hdd1/llms, all text; no ECAPA, no
x-vector, no titanet, no wespeaker, no .onnx anywhere under /mnt/hdd1. So
newSpeakerEmbedder returns nil, internal/speaker falls back to
speaker.Disabled, Identify answers ErrDisabled, and the daemon logs which
half is off at startup. The plan's "simple MFCC + GMM" floor is refused in
the package comment: MFCC cosine distance detects channel and loudness as
much as voice, and a biometric that is confidently wrong writes false claims
about named people into his memory. A bad floor is worse than none here.

Refused as well, and the reason is in enroll.go's doc comment: the plan asked
for unknown speakers to be enrolled on first interaction with a TTS "кто
это?". There is no request shape in the protocol that could express that.
Taking a biometric of whoever walks past the microphone does it to guests who
are not party to the exchange, and a synthesised question into a room is not
consent from whoever answers.

Authority: enrolment is AuthStepUp, because it is a deliberate sit-down act
that writes a biometric of a named person and never something done by voice
mid-conversation. Deletion is one rung lower at AuthWrite, deliberately
inverting the usual pattern — getting rid of a biometric must never be the
harder half. Listing is AuthRead and never returns the vectors themselves.

Off unless configured: no speaker block means the three methods answer
ErrUnknownMethod, so a default box has no wire path that takes a voiceprint.

make build and make test pass.

Vikunja #255

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TrVSBKe3RFDF4fGYKWYQnX
2026-08-01 05:23:03 +04:00

244 lines
11 KiB
Go

package auth
import (
"encoding/json"
"errors"
"fmt"
"github.com/kami/maven/internal/ipc"
)
// Authority — a discrete authority requirement per ipc.Method. Higher
// numbers are STRICTER (need a higher layer to be granted). Today's CoreAPI
// methods are all read or single-module-write; the deferred L3 acts
// (EnableTool / destructive Ops) are reserved at the top rung — they're
// not on the CoreAPI yet (the tool-executor module is unbuilt), but the
// authority table holds the rung so adding them is a policy entry, not a
// new mechanism.
type Authority int8
const (
// AuthRead — read methods (LatestFact, LatestFactBySource, Since, Presence,
// RecentOutcomes) and state mutations a module legitimately makes
// (CreateReminder, MarkReminder, RecordNudge, ResolveNudge). The Enrollment
// already gated caller identity; any enrolled module may use these.
AuthRead Authority = 0
// AuthWrite — WriteFact. Need enrollment + source-scope match. The
// "compromised poller can't forge a trigger" property: a module only writes
// sources it owns. Floor gets "*"; tight enrollments scope per source.
AuthWrite Authority = 1
// AuthStepUp — a per-assertion user-verification gesture is required for
// this call. Reserved for EnableTool (registration-enable) and destructive
// acts when they land on the CoreAPI. NOT a Layer itself — Authority is
// the call-side requirement; Layer is the surface-side capability. The
// pure check is just: does the surface cap (MaxLayer) carry L3, AND was
// step-up asserted this session? Both settled by Can below.
AuthStepUp Authority = 2
)
// Requirement — the PURE authority table: per-method required Authority. This
// is the one place in the codebase a method's required authority is declared;
// every other reference to "registration needs step-up" points back here.
// Adding a new ipc.Method = a row here (or it inherits AuthRead by default,
// which the vet check in dispatch catches via Method existence, not auth).
func Requirement(m ipc.Method) Authority {
switch m {
case ipc.MethodEnableTool, ipc.MethodDisableTool:
// Both mutate the tool allowlist — the boundary. Enable adds a runnable
// capability (privilege escalation); disable removes one (fail-safe
// direction, but still an allowlist mutation and a lever an attacker
// could pull to silence a security-relevant tool). Human-only, step-up
// asserted — never a module or the voice/chat path. maven can propose
// (MethodProposeTool, no step-up: she has no passkey) but never en/disable.
return AuthStepUp
case ipc.MethodSwapModel:
// Swapping the resident model changes what routes every utterance and
// what words every reply. It is the owner's call, from a surface that can
// carry a passkey gesture — the same rung as mutating the tool allowlist,
// and for the same reason: nothing Maven says or does may reach it.
// MethodModelStatus is only the read side, so it stays at AuthRead.
return AuthStepUp
case ipc.MethodCaptureStart, ipc.MethodCaptureAppend, ipc.MethodCaptureStop:
// Recording a meeting (Vikunja #253). AuthWrite, not AuthRead: it puts
// audio of other people on disk, which is a heavier thing than reading a
// fact, and it is not something a read-only surface should be able to
// begin. Append and Stop sit on the same rung as Start deliberately —
// a surface that may not start a recording has no business feeding or
// harvesting one either.
//
// Not AuthStepUp, and this is the interesting line: step-up needs a
// passkey gesture, which the voice path cannot make. Putting it here
// would mean "запиши встречу" could never work by voice, and the real
// gate on this capability is elsewhere and stronger — the methods do not
// exist at all unless the operator enabled a capture block, and no
// recording can begin without someone saying so.
return AuthWrite
case ipc.MethodEnrollSpeaker:
// Taking a voiceprint (Vikunja #255). AuthStepUp, and unlike recording a
// meeting there is no reason to soften it: enrolment is not a thing anyone
// does by voice mid-conversation. It is a deliberate sit-down with a
// surface that can carry a passkey gesture, and it writes a biometric of a
// named person. If the gesture is inconvenient, that is the correct amount
// of friction for this particular write.
return AuthStepUp
case ipc.MethodForgetSpeaker:
// Deleting a voiceprint. One rung BELOW enrolment on purpose. Everywhere
// else in this table the destructive direction is gated at least as hard
// as the constructive one, and here that would be wrong: getting rid of a
// biometric must never be the harder half. The worst a caller at this rung
// can do is make Maven stop recognising someone, which is the state the
// box ships in anyway.
return AuthWrite
case ipc.MethodWriteFact:
return AuthWrite
case ipc.MethodAssertStepUp:
return AuthRead
case ipc.MethodLatestFact,
ipc.MethodLatestFactBySource,
ipc.MethodSince,
ipc.MethodPresence,
ipc.MethodRecentOutcomes,
ipc.MethodCreateReminder,
ipc.MethodMarkReminder,
ipc.MethodRecordNudge,
ipc.MethodResolveNudge,
// Task capture (Vikunja #130). Listed explicitly rather than left to
// the default so the intent is on the record: capturing a task is a
// module write, not an allowlist mutation and not a new standing reason
// for Maven to speak — nothing in the tick loop reads tasks. It stays
// at AuthRead, the same rung as CreateReminder, which is the closest
// existing analogue.
ipc.MethodCaptureTask,
ipc.MethodListTasks,
ipc.MethodSetTaskStatus,
// Mail ingestion (Vikunja #246). AuthRead because of what the method can
// produce: candidate tasks and nothing else. It cannot write a fact, set a
// reminder, or touch the tool allowlist, so a compromised mail reader can
// at worst put junk on a review page he clears in one click.
ipc.MethodIngestMail,
// Looking at one image (Vikunja #252). AuthRead because of what it can
// produce: words about a picture, and optionally a note. It cannot write
// a fact, set a reminder, or touch the tool allowlist. The invasive part
// of this capability is not the authority rung — it is that the bytes are
// kept on disk, which media.retention bounds, and that they never leave
// the box, which internal/vision enforces by refusing a non-private
// endpoint.
ipc.MethodDescribeImage,
// "что ты записываешь?" — the read side of the recorder. It reports a
// label, a start time and a byte count, begins nothing and keeps nothing.
ipc.MethodCaptureStatus,
// Who is enrolled. Returns ids, names and enrolment dates — never the
// voiceprints themselves, which stay in core. Listing the people Maven can
// recognise is exactly the read a surface needs to offer a "forget" button.
ipc.MethodListSpeakers,
// The read side of the model swap: which model is resident, which ones are
// allowlisted. It loads nothing and changes nothing.
ipc.MethodModelStatus:
return AuthRead
}
// Unknown method ⇒ AuthRead, but ipc.dispatch returns ErrUnknownMethod
// regardless of the auth verdict (we run before dispatch; we don't gate on
// Method existence — Check is method-agnostic policy, not routing).
return AuthRead
}
// Can — the PURE authority decision for one call. Returns nil if the scope
// is authorized to invoke m with the supplied params; an error otherwise:
//
// - ErrUnenrolled — scope has no Module (caller wasn't in the enrollment
// table). Fail closed.
// - ErrForbidden — surface caps the layer below what m requires, or
// WriteFact's source is out of scope.
//
// The adapter wrapping this for the ipc gate (Gate.Check) maps the error to
// codeForbidden at the wire; we keep the distinction here so daemon logs
// can show why a call was denied.
//
// params is the raw json.RawMessage the ipc Server received; for WriteFact we
// re-parse Source out of it. Other methods don't need params — their verdict
// depends only on the scope.
func Can(m ipc.Method, scope Scope, params json.RawMessage) error {
// Floor closed: any caller not in the enrollment table is refused outright.
// Not "0-level unauthed" — refused. This is the surface-caps property
// applied before the layer caps: there is no L0 surface if SurfaceUnknown.
if scope.Module == "" {
return ErrUnenrolled
}
if scope.Surface == SurfaceUnknown {
return ErrUnenrolled
}
switch Requirement(m) {
case AuthRead:
// Any enrolled module may read. Reads through the surface level the
// Enrollment set (voice-L0 wouldn't be enrolled to write at all).
return nil
case AuthWrite:
if m == ipc.MethodWriteFact {
src, err := extractSource(params)
if err != nil {
// Malformed params is a bad-params error already produced by
// ipc.dispatch; but Can runs first. Treat as forbidden — a
// caller doesn't get to probe scopes with garbage params.
return fmt.Errorf("%w: malformed source", ErrForbidden)
}
if !SourceAllowed(scope.SourceScope, src) {
// The spec's compromised-poller case in one line: a poller
// enrolled to write poll:healthcheck asking to write
// poll:uptime is denied — but the same poller writing
// poll:healthcheck is fine. Polls can't forge triggers.
return fmt.Errorf("%w: source %q out of scope", ErrForbidden, src)
}
}
return nil
case AuthStepUp:
// Surface-caps-authority enforced here. The surface can't carry L3 ⇒
// forbidden. The session step-up itself is checked by the Gate (it
// owns Session state and surfaces a Check function); we cap surface
// here so the gate fails closed on shape alone.
if MaxLayer(scope.Surface) < Layer3 {
return fmt.Errorf("%w: surface %s can't carry step-up", ErrForbidden, scope.Surface)
}
return nil
}
return nil
}
// SourceAllowed — true iff src is in scope (the wildcard "*" matches all).
// Empty scope ⇒ fail closed. The function is pure; we keep it exported so a
// future enrollment table can call into the same matching logic.
func SourceAllowed(scope []string, src string) bool {
if len(scope) == 0 {
return false
}
for _, s := range scope {
if s == "*" || s == src {
return true
}
}
return false
}
// extractSource reads WriteFactReq.Source out of the raw params WITHOUT a full
// unmarshal — Source is the only field Can needs, and re-parsing it once per
// write is cheap (and only happens on MethodWriteFact). Stay independent of
// any future WriteFactReq shape changes by using the struct directly.
func extractSource(raw json.RawMessage) (string, error) {
var p ipc.WriteFactReq
if err := json.Unmarshal(raw, &p); err != nil {
return "", err
}
if p.Source == "" {
// An empty source is rejected by store.WriteFact anyway; surface it
// as forbidden to avoid giving a caller an ipc sentinel that names the
// store's internal invariant. (store rejects this before feature flag
// for "missing source"; today this is best-effort.)
return "", errors.New("empty source")
}
return p.Source, nil
}