Files
openclaw/docs/plugins/sdk-channel-ingress.md
Josh Avant 97a53a9b35 feat: audit admitted channel participant identity (#122863)
* feat: audit admitted channel participant identity

* fix: preserve Telegram identity through thread recovery

* fix: signal held gateway process groups

* fix: keep audit evidence passive in collect routing

* fix: validate copied channel participant evidence

* fix: bind channel participant evidence to host ingress

* fix: honor Telegram proof credential roles

* fix: restart held Telegram proof through gateway

* fix: repair channel identity CI regressions

* test(matrix): bind thread routing owner

* fix: preserve direct DM SDK compatibility

* fix: bind channel provenance at host runtime

* test(feishu): provide channel context builder

* fix: defer record-bound channel runtime resolution

* fix: keep channel admission evidence core-private

* fix(audit): bind channel admission to plugin lifecycle

* fix(audit): bind ingress provenance to final context

* refactor(audit): split admission scope keys

* test(queue): cover combined metadata carriers

* refactor(audit): keep lifecycle helpers private

* fix(queue): preserve combined turn authority

* test(channels): provide ingress context builders

* test(channels): align integrated CI fixtures

* test(clickclack): resolve model-loop ingress

* docs: preserve channel participant evidence invariant
2026-08-14 08:57:01 -05:00

8.2 KiB

summary, read_when, title, sidebarTitle
summary read_when title sidebarTitle
Experimental channel ingress API for inbound message authorization
Building or migrating a messaging channel plugin
Changing DM or group allowlists, route gates, command auth, event auth, or mention activation
Reviewing channel ingress redaction or SDK compatibility boundaries
Channel ingress API Channel Ingress

Channel ingress is the experimental access-control boundary for inbound channel events. Plugins own platform facts and side effects; core owns generic policy: DM/group allowlists, pairing-store DM entries, route gates, command gates, event auth, mention activation, redacted diagnostics, and admission.

Use openclaw/plugin-sdk/channel-ingress-runtime for receive paths.

Runtime resolver

import {
  defineStableChannelIngressIdentity,
  resolveChannelMessageIngress,
} from "openclaw/plugin-sdk/channel-ingress-runtime";

const identity = defineStableChannelIngressIdentity({
  key: "platform-user-id",
  normalize: normalizePlatformUserId,
  sensitivity: "pii",
});

const result = await resolveChannelMessageIngress({
  channelId: "my-channel",
  accountId,
  identity,
  subject: { stableId: platformUserId },
  conversation: { kind: isGroup ? "group" : "direct", id: conversationId },
  contextBinding: {
    agentId: agentRoute.agentId,
    sessionKey: agentRoute.sessionKey,
    messageId,
    inboundEventKind: "user_request",
  },
  event: { kind: "message", authMode: "inbound", mayPair: !isGroup },
  policy: {
    dmPolicy: config.dmPolicy,
    groupPolicy: config.groupPolicy,
    groupAllowFromFallbackToAllowFrom: true,
  },
  allowFrom: config.allowFrom,
  groupAllowFrom: config.groupAllowFrom,
  accessGroups: cfg.accessGroups,
  route,
  readStoreAllowFrom,
  command: hasControlCommand ? { allowTextCommands: true, hasControlCommand } : undefined,
});

const ctx = runtime.channel.inbound.buildContext({
  // Pass the exact host result; do not rebuild participant evidence from
  // SenderId, From, session keys, routes, rooms, or message metadata.
  channelIngress: result,
  // ...normalized channel facts
});

Do not precompute effective allowlists, command owners, or command groups. The resolver derives them from raw allowlists, store callbacks, route descriptors, access groups, policy, and conversation kind.

For a result that will enter a host context, resolve after the channel's route owner has selected the final agent and session. contextBinding freezes those facts with the stable transport message id (when present) and final inbound event kind. Decision-only checks may omit it, but such a result is not valid execution provenance and must not be passed as channelIngress. When a channel batches several admitted messages, pass their exact results in source order; the finalized context message id identifies the last source result.

Result

Bundled plugins should consume modern projections directly:

Field Meaning
ingress ordered gate decision and admission
senderAccess sender/conversation authorization only
routeAccess route and route-sender projection
commandAccess command authorization; requested: false when no command gate ran
activationAccess mention/activation result

Event authorization stays available on the ordered ingress.graph and the decisive ingress.reasonCode; no separate event projection is emitted.

Deprecated third-party SDK helpers may rebuild older shapes internally. New bundled receive paths should not translate modern results back into local DTOs.

When execution-identity audit collection is enabled, a trusted active native plugin is the authoritative in-process producer of its remote participant fact. The host-injected registered runtime binds the resolver result to the exact plugin record and registry lifecycle epoch, then validates its complete available conversation, route, agent, session, message, event, and participant scope during a one-shot context handoff. The public standalone builder remains non-authoritative and cannot mint participant evidence. Queue collection retains attribution only when every contribution has valid evidence for the same participant; mixed, missing, stale, or unminted evidence is unknown. The carrier is opaque, bounded, one-shot, and diagnostic only. Plugins cannot mint participant evidence from caller-chosen sender, account, room, route, session, message, or transport fields. The SDK intentionally exposes no record, epoch, owner capability, participant-evidence constructor, or evidence copier. A structurally similar result, stale record, reused result, or scope-changed context does not gain host authority.

boundary-verified means core verified that the participant fact crossed this trusted active registered native-plugin boundary with the exact record, epoch, scope, and one-shot handoff. It does not mean core independently queried the remote service; only the channel plugin can observe that transport fact.

The audit states are distinct:

  • supported: the authoritative ingress resolver ran. Its exact result can yield a present invoker and enforced or attribution-only coverage.
  • unknown: a supported handoff was missing, stale, fake, reused, mixed, or otherwise failed host validation. Unknown never means allowed.
  • unsupported: a named path has no Phase 0 authoritative integration and explicitly passes channelIngress: "unsupported". Unsupported never means allowed and is not a shortcut for incomplete wiring.

Access groups

accessGroup:<name> entries stay redacted. Core resolves static message.senders groups itself and calls resolveAccessGroupMembership only for dynamic groups that require a platform lookup. Missing, unsupported, and failed groups fail closed.

Event modes

authMode Meaning
inbound normal inbound sender gates
command command gates for callbacks or scoped buttons
origin-subject actor must match the original message subject
route-only route gates only for route-scoped trusted events
none plugin-owned internal events bypass shared auth

Use mayPair: false for reactions, buttons, callbacks, and native commands.

Routes and activation

Use route descriptors for room, topic, guild, thread, or nested route policy:

route: {
  id: "room",
  allowed: roomAllowed,
  enabled: roomEnabled,
  senderPolicy: "replace",
  senderAllowFrom: roomAllowFrom,
  blockReason: "room_sender_not_allowlisted",
}

Use channelIngressRoutes(...) when a plugin has several optional route descriptors; it filters disabled branches while keeping route facts generic and ordered by each descriptor's precedence.

Mention gating is an activation gate. A mention miss returns admission: "skip" so the turn kernel does not process an observe-only turn. Most channels should leave activation after sender and command gates. Public chat surfaces that must quiet non-mentioned traffic before sender allowlist noise can opt into activation.order: "before-sender" when text-command bypass is disabled. Channels with implicit activation, such as replies in bot threads, resolve channels.defaults.implicitMentions plus channel and account overrides with resolveChannelImplicitMentions(...), then pass the result as activation.implicitMentions. The projected activationAccess.shouldBypassMention reports when command or implicit activation bypassed an explicit mention.

Redaction

Raw sender values and raw allowlist entries are resolver input only. They must not appear in resolved state, decisions, diagnostics, snapshots, or compatibility facts. Use opaque subject ids, entry ids, route ids, and diagnostic ids.

Verification

pnpm test src/channels/message-access/message-access.test.ts src/plugin-sdk/channel-ingress-runtime.test.ts
pnpm plugin-sdk:api:diff --base "$(git merge-base origin/main HEAD)" --head HEAD