* refactor: introduce provider plugin registry * refactor: move provider CLI to plugins * docs: add provider plugin implementation notes * refactor: shift provider runtime logic into plugins * refactor: add plugin defaults and summaries * docs: update provider plugin notes * feat(commands): add /commands slash list * Auto-reply: tidy help message * Auto-reply: fix status command lint * Tests: align google shared expectations * Auto-reply: tidy help message * Auto-reply: fix status command lint * refactor: move provider routing into plugins * test: align agent routing expectations * docs: update provider plugin notes * refactor: route replies via provider plugins * docs: note route-reply plugin hooks * refactor: extend provider plugin contract * refactor: derive provider status from plugins * refactor: unify gateway provider control * refactor: use plugin metadata in auto-reply * fix: parenthesize cron target selection * refactor: derive gateway methods from plugins * refactor: generalize provider logout * refactor: route provider logout through plugins * refactor: move WhatsApp web login methods into plugin * refactor: generalize provider log prefixes * refactor: centralize default chat provider * refactor: derive provider lists from registry * refactor: move provider reload noops into plugins * refactor: resolve web login provider via alias * refactor: derive CLI provider options from plugins * refactor: derive prompt provider list from plugins * style: apply biome lint fixes * fix: resolve provider routing edge cases * docs: update provider plugin refactor notes * fix(gateway): harden agent provider routing * refactor: move provider routing into plugins * refactor: move provider CLI to plugins * refactor: derive provider lists from registry * fix: restore slash command parsing * refactor: align provider ids for schema * refactor: unify outbound target resolution * fix: keep outbound labels stable * feat: add msteams to cron surfaces * fix: clean up lint build issues * refactor: localize chat provider alias normalization * refactor: drive gateway provider lists from plugins * docs: update provider plugin notes * style: format message-provider * fix: avoid provider registry init cycles * style: sort message-provider imports * fix: relax provider alias map typing * refactor: move provider routing into plugins * refactor: add plugin pairing/config adapters * refactor: route pairing and provider removal via plugins * refactor: align auto-reply provider typing * test: stabilize telegram media mocks * docs: update provider plugin refactor notes * refactor: pluginize outbound targets * refactor: pluginize provider selection * refactor: generalize text chunk limits * docs: update provider plugin notes * refactor: generalize group session/config * fix: normalize provider id for room detection * fix: avoid provider init in system prompt * style: formatting cleanup * refactor: normalize agent delivery targets * test: update outbound delivery labels * chore: fix lint regressions * refactor: extend provider plugin adapters * refactor: move elevated/block streaming defaults to plugins * refactor: defer outbound send deps to plugins * docs: note plugin-driven streaming/elevated defaults * refactor: centralize webchat provider constant * refactor: add provider setup adapters * refactor: delegate provider add config to plugins * docs: document plugin-driven provider add * refactor: add plugin state/binding metadata * refactor: build agent provider status from plugins * docs: note plugin-driven agent bindings * refactor: centralize internal provider constant usage * fix: normalize WhatsApp targets for groups and E.164 (#631) (thanks @imfing) * refactor: centralize default chat provider * refactor: centralize WhatsApp target normalization * refactor: move provider routing into plugins * refactor: normalize agent delivery targets * chore: fix lint regressions * fix: normalize WhatsApp targets for groups and E.164 (#631) (thanks @imfing) * feat: expand provider plugin adapters * refactor: route auto-reply via provider plugins * fix: align WhatsApp target normalization * fix: normalize WhatsApp targets for groups and E.164 (#631) (thanks @imfing) * refactor: centralize WhatsApp target normalization * feat: add /config chat config updates * docs: add /config get alias * feat(commands): add /commands slash list * refactor: centralize default chat provider * style: apply biome lint fixes * chore: fix lint regressions * fix: clean up whatsapp allowlist typing * style: format config command helpers * refactor: pluginize tool threading context * refactor: normalize session announce targets * docs: note new plugin threading and announce hooks * refactor: pluginize message actions * docs: update provider plugin actions notes * fix: align provider action adapters * refactor: centralize webchat checks * style: format message provider helpers * refactor: move provider onboarding into adapters * docs: note onboarding provider adapters * feat: add msteams onboarding adapter * style: organize onboarding imports * fix: normalize msteams allowFrom types * feat: add plugin text chunk limits * refactor: use plugin chunk limit fallbacks * feat: add provider mention stripping hooks * style: organize provider plugin type imports * refactor: generalize health snapshots * refactor: update macOS health snapshot handling * docs: refresh health snapshot notes * style: format health snapshot updates * refactor: drive security warnings via plugins * docs: note provider security adapter * style: format provider security adapters * refactor: centralize provider account defaults * refactor: type gateway client identity constants * chore: regen gateway protocol swift * fix: degrade health on failed provider probe * refactor: centralize pairing approve hint * docs: add plugin CLI command references * refactor: route auth and tool sends through plugins * docs: expand provider plugin hooks * refactor: document provider docking touchpoints * refactor: normalize internal provider defaults * refactor: streamline outbound delivery wiring * refactor: make provider onboarding plugin-owned * refactor: support provider-owned agent tools * refactor: move telegram draft chunking into telegram module * refactor: infer provider tool sends via extractToolSend * fix: repair plugin onboarding imports * refactor: de-dup outbound target normalization * style: tidy plugin and agent imports * refactor: data-drive provider selection line * fix: satisfy lint after provider plugin rebase * test: deflake gateway-cli coverage * style: format gateway-cli coverage test * refactor(provider-plugins): simplify provider ids * test(pairing-cli): avoid provider-specific ternary * style(macos): swiftformat HealthStore * refactor(sandbox): derive provider tool denylist * fix(sandbox): avoid plugin init in defaults * refactor(provider-plugins): centralize provider aliases * style(test): satisfy biome * refactor(protocol): v3 providers.status maps * refactor(ui): adapt to protocol v3 * refactor(macos): adapt to protocol v3 * test: update providers.status v3 fixtures * refactor(gateway): map provider runtime snapshot * test(gateway): update reload runtime snapshot * refactor(whatsapp): normalize heartbeat provider id * docs(refactor): update provider plugin notes * style: satisfy biome after rebase * fix: describe sandboxed elevated in prompt * feat(gateway): add agent image attachments + live probe * refactor: derive CLI provider options from plugins * fix(gateway): harden agent provider routing * fix(gateway): harden agent provider routing * refactor: align provider ids for schema * fix(protocol): keep agent provider string * fix(gateway): harden agent provider routing * fix(protocol): keep agent provider string * refactor: normalize agent delivery targets * refactor: support provider-owned agent tools * refactor(config): provider-keyed elevated allowFrom * style: satisfy biome * fix(gateway): appease provider narrowing * style: satisfy biome * refactor(reply): move group intro hints into plugin * fix(reply): avoid plugin registry init cycle * refactor(providers): add lightweight provider dock * refactor(gateway): use typed client id in connect * refactor(providers): document docks and avoid init cycles * refactor(providers): make media limit helper generic * fix(providers): break plugin registry import cycles * style: satisfy biome * refactor(status-all): build providers table from plugins * refactor(gateway): delegate web login to provider plugin * refactor(provider): drop web alias * refactor(provider): lazy-load monitors * style: satisfy lint/format * style: format status-all providers table * style: swiftformat gateway discovery model * test: make reload plan plugin-driven * fix: avoid token stringification in status-all * refactor: make provider IDs explicit in status * feat: warn on signal/imessage provider runtime errors * test: cover gateway provider runtime warnings in status * fix: add runtime kind to provider status issues * test: cover health degradation on probe failure * fix: keep routeReply lightweight * style: organize routeReply imports * refactor(web): extract auth-store helpers * refactor(whatsapp): lazy login imports * refactor(outbound): route replies via plugin outbound * docs: update provider plugin notes * style: format provider status issues * fix: make sandbox scope warning wrap-safe * refactor: load outbound adapters from provider plugins * docs: update provider plugin outbound notes * style(macos): fix swiftformat lint * docs: changelog for provider plugins * fix(macos): satisfy swiftformat * fix(macos): open settings via menu action * style: format after rebase * fix(macos): open Settings via menu action --------- Co-authored-by: LK <luke@kyohere.com> Co-authored-by: Luke K (pr-0f3t) <2609441+lc0rp@users.noreply.github.com> Co-authored-by: Xin <xin@imfing.com>
305 lines
11 KiB
TypeScript
305 lines
11 KiB
TypeScript
import crypto from "node:crypto";
|
|
import { lookupContextTokens } from "../../agents/context.js";
|
|
import { DEFAULT_CONTEXT_TOKENS } from "../../agents/defaults.js";
|
|
import { runWithModelFallback } from "../../agents/model-fallback.js";
|
|
import { runEmbeddedPiAgent } from "../../agents/pi-embedded.js";
|
|
import { hasNonzeroUsage } from "../../agents/usage.js";
|
|
import {
|
|
type SessionEntry,
|
|
updateSessionStoreEntry,
|
|
} from "../../config/sessions.js";
|
|
import type { TypingMode } from "../../config/types.js";
|
|
import { logVerbose } from "../../globals.js";
|
|
import { registerAgentRunContext } from "../../infra/agent-events.js";
|
|
import { defaultRuntime } from "../../runtime.js";
|
|
import { stripHeartbeatToken } from "../heartbeat.js";
|
|
import type { OriginatingChannelType } from "../templating.js";
|
|
import { isSilentReplyText, SILENT_REPLY_TOKEN } from "../tokens.js";
|
|
import type { GetReplyOptions, ReplyPayload } from "../types.js";
|
|
import type { FollowupRun } from "./queue.js";
|
|
import {
|
|
applyReplyThreading,
|
|
filterMessagingToolDuplicates,
|
|
shouldSuppressMessagingToolReplies,
|
|
} from "./reply-payloads.js";
|
|
import { resolveReplyToMode } from "./reply-threading.js";
|
|
import { isRoutableChannel, routeReply } from "./route-reply.js";
|
|
import { incrementCompactionCount } from "./session-updates.js";
|
|
import type { TypingController } from "./typing.js";
|
|
import { createTypingSignaler } from "./typing-mode.js";
|
|
|
|
export function createFollowupRunner(params: {
|
|
opts?: GetReplyOptions;
|
|
typing: TypingController;
|
|
typingMode: TypingMode;
|
|
sessionEntry?: SessionEntry;
|
|
sessionStore?: Record<string, SessionEntry>;
|
|
sessionKey?: string;
|
|
storePath?: string;
|
|
defaultModel: string;
|
|
agentCfgContextTokens?: number;
|
|
}): (queued: FollowupRun) => Promise<void> {
|
|
const {
|
|
opts,
|
|
typing,
|
|
typingMode,
|
|
sessionEntry,
|
|
sessionStore,
|
|
sessionKey,
|
|
storePath,
|
|
defaultModel,
|
|
agentCfgContextTokens,
|
|
} = params;
|
|
const typingSignals = createTypingSignaler({
|
|
typing,
|
|
mode: typingMode,
|
|
isHeartbeat: opts?.isHeartbeat === true,
|
|
});
|
|
|
|
/**
|
|
* Sends followup payloads, routing to the originating channel if set.
|
|
*
|
|
* When originatingChannel/originatingTo are set on the queued run,
|
|
* replies are routed directly to that provider instead of using the
|
|
* session's current dispatcher. This ensures replies go back to
|
|
* where the message originated.
|
|
*/
|
|
const sendFollowupPayloads = async (
|
|
payloads: ReplyPayload[],
|
|
queued: FollowupRun,
|
|
) => {
|
|
// Check if we should route to originating channel.
|
|
const { originatingChannel, originatingTo } = queued;
|
|
const shouldRouteToOriginating =
|
|
isRoutableChannel(originatingChannel) && originatingTo;
|
|
|
|
if (!shouldRouteToOriginating && !opts?.onBlockReply) {
|
|
logVerbose("followup queue: no onBlockReply handler; dropping payloads");
|
|
return;
|
|
}
|
|
|
|
for (const payload of payloads) {
|
|
if (!payload?.text && !payload?.mediaUrl && !payload?.mediaUrls?.length) {
|
|
continue;
|
|
}
|
|
if (
|
|
isSilentReplyText(payload.text, SILENT_REPLY_TOKEN) &&
|
|
!payload.mediaUrl &&
|
|
!payload.mediaUrls?.length
|
|
) {
|
|
continue;
|
|
}
|
|
await typingSignals.signalTextDelta(payload.text);
|
|
|
|
// Route to originating channel if set, otherwise fall back to dispatcher.
|
|
if (shouldRouteToOriginating) {
|
|
const result = await routeReply({
|
|
payload,
|
|
channel: originatingChannel,
|
|
to: originatingTo,
|
|
sessionKey: queued.run.sessionKey,
|
|
accountId: queued.originatingAccountId,
|
|
threadId: queued.originatingThreadId,
|
|
cfg: queued.run.config,
|
|
});
|
|
if (!result.ok) {
|
|
// Log error and fall back to dispatcher if available.
|
|
const errorMsg = result.error ?? "unknown error";
|
|
logVerbose(`followup queue: route-reply failed: ${errorMsg}`);
|
|
// Fallback: try the dispatcher if routing failed.
|
|
if (opts?.onBlockReply) {
|
|
await opts.onBlockReply(payload);
|
|
}
|
|
}
|
|
} else if (opts?.onBlockReply) {
|
|
await opts.onBlockReply(payload);
|
|
}
|
|
}
|
|
};
|
|
|
|
return async (queued: FollowupRun) => {
|
|
await typingSignals.signalRunStart();
|
|
try {
|
|
const runId = crypto.randomUUID();
|
|
if (queued.run.sessionKey) {
|
|
registerAgentRunContext(runId, {
|
|
sessionKey: queued.run.sessionKey,
|
|
verboseLevel: queued.run.verboseLevel,
|
|
});
|
|
}
|
|
let autoCompactionCompleted = false;
|
|
let runResult: Awaited<ReturnType<typeof runEmbeddedPiAgent>>;
|
|
let fallbackProvider = queued.run.provider;
|
|
let fallbackModel = queued.run.model;
|
|
try {
|
|
const fallbackResult = await runWithModelFallback({
|
|
cfg: queued.run.config,
|
|
provider: queued.run.provider,
|
|
model: queued.run.model,
|
|
run: (provider, model) =>
|
|
runEmbeddedPiAgent({
|
|
sessionId: queued.run.sessionId,
|
|
sessionKey: queued.run.sessionKey,
|
|
messageProvider: queued.run.messageProvider,
|
|
agentAccountId: queued.run.agentAccountId,
|
|
sessionFile: queued.run.sessionFile,
|
|
workspaceDir: queued.run.workspaceDir,
|
|
config: queued.run.config,
|
|
skillsSnapshot: queued.run.skillsSnapshot,
|
|
prompt: queued.prompt,
|
|
extraSystemPrompt: queued.run.extraSystemPrompt,
|
|
ownerNumbers: queued.run.ownerNumbers,
|
|
enforceFinalTag: queued.run.enforceFinalTag,
|
|
provider,
|
|
model,
|
|
authProfileId: queued.run.authProfileId,
|
|
thinkLevel: queued.run.thinkLevel,
|
|
verboseLevel: queued.run.verboseLevel,
|
|
reasoningLevel: queued.run.reasoningLevel,
|
|
bashElevated: queued.run.bashElevated,
|
|
timeoutMs: queued.run.timeoutMs,
|
|
runId,
|
|
blockReplyBreak: queued.run.blockReplyBreak,
|
|
onAgentEvent: (evt) => {
|
|
if (evt.stream !== "compaction") return;
|
|
const phase =
|
|
typeof evt.data.phase === "string" ? evt.data.phase : "";
|
|
const willRetry = Boolean(evt.data.willRetry);
|
|
if (phase === "end" && !willRetry) {
|
|
autoCompactionCompleted = true;
|
|
}
|
|
},
|
|
}),
|
|
});
|
|
runResult = fallbackResult.result;
|
|
fallbackProvider = fallbackResult.provider;
|
|
fallbackModel = fallbackResult.model;
|
|
} catch (err) {
|
|
const message = err instanceof Error ? err.message : String(err);
|
|
defaultRuntime.error?.(
|
|
`Followup agent failed before reply: ${message}`,
|
|
);
|
|
return;
|
|
}
|
|
|
|
const payloadArray = runResult.payloads ?? [];
|
|
if (payloadArray.length === 0) return;
|
|
const sanitizedPayloads = payloadArray.flatMap((payload) => {
|
|
const text = payload.text;
|
|
if (!text || !text.includes("HEARTBEAT_OK")) return [payload];
|
|
const stripped = stripHeartbeatToken(text, { mode: "message" });
|
|
const hasMedia =
|
|
Boolean(payload.mediaUrl) || (payload.mediaUrls?.length ?? 0) > 0;
|
|
if (stripped.shouldSkip && !hasMedia) return [];
|
|
return [{ ...payload, text: stripped.text }];
|
|
});
|
|
const replyToChannel =
|
|
queued.originatingChannel ??
|
|
(queued.run.messageProvider?.toLowerCase() as
|
|
| OriginatingChannelType
|
|
| undefined);
|
|
const replyToMode = resolveReplyToMode(
|
|
queued.run.config,
|
|
replyToChannel,
|
|
queued.originatingAccountId,
|
|
);
|
|
|
|
const replyTaggedPayloads: ReplyPayload[] = applyReplyThreading({
|
|
payloads: sanitizedPayloads,
|
|
replyToMode,
|
|
replyToChannel,
|
|
});
|
|
|
|
const dedupedPayloads = filterMessagingToolDuplicates({
|
|
payloads: replyTaggedPayloads,
|
|
sentTexts: runResult.messagingToolSentTexts ?? [],
|
|
});
|
|
const suppressMessagingToolReplies = shouldSuppressMessagingToolReplies({
|
|
messageProvider: queued.run.messageProvider,
|
|
messagingToolSentTargets: runResult.messagingToolSentTargets,
|
|
originatingTo: queued.originatingTo,
|
|
accountId: queued.run.agentAccountId,
|
|
});
|
|
const finalPayloads = suppressMessagingToolReplies ? [] : dedupedPayloads;
|
|
|
|
if (finalPayloads.length === 0) return;
|
|
|
|
if (autoCompactionCompleted) {
|
|
const count = await incrementCompactionCount({
|
|
sessionEntry,
|
|
sessionStore,
|
|
sessionKey,
|
|
storePath,
|
|
});
|
|
if (queued.run.verboseLevel === "on") {
|
|
const suffix = typeof count === "number" ? ` (count ${count})` : "";
|
|
finalPayloads.unshift({
|
|
text: `🧹 Auto-compaction complete${suffix}.`,
|
|
});
|
|
}
|
|
}
|
|
|
|
if (storePath && sessionKey) {
|
|
const usage = runResult.meta.agentMeta?.usage;
|
|
const modelUsed =
|
|
runResult.meta.agentMeta?.model ?? fallbackModel ?? defaultModel;
|
|
const contextTokensUsed =
|
|
agentCfgContextTokens ??
|
|
lookupContextTokens(modelUsed) ??
|
|
sessionEntry?.contextTokens ??
|
|
DEFAULT_CONTEXT_TOKENS;
|
|
|
|
if (hasNonzeroUsage(usage)) {
|
|
try {
|
|
await updateSessionStoreEntry({
|
|
storePath,
|
|
sessionKey,
|
|
update: async (entry) => {
|
|
const input = usage.input ?? 0;
|
|
const output = usage.output ?? 0;
|
|
const promptTokens =
|
|
input + (usage.cacheRead ?? 0) + (usage.cacheWrite ?? 0);
|
|
return {
|
|
inputTokens: input,
|
|
outputTokens: output,
|
|
totalTokens:
|
|
promptTokens > 0 ? promptTokens : (usage.total ?? input),
|
|
modelProvider: fallbackProvider ?? entry.modelProvider,
|
|
model: modelUsed,
|
|
contextTokens: contextTokensUsed ?? entry.contextTokens,
|
|
updatedAt: Date.now(),
|
|
};
|
|
},
|
|
});
|
|
} catch (err) {
|
|
logVerbose(
|
|
`failed to persist followup usage update: ${String(err)}`,
|
|
);
|
|
}
|
|
} else if (modelUsed || contextTokensUsed) {
|
|
try {
|
|
await updateSessionStoreEntry({
|
|
storePath,
|
|
sessionKey,
|
|
update: async (entry) => ({
|
|
modelProvider: fallbackProvider ?? entry.modelProvider,
|
|
model: modelUsed ?? entry.model,
|
|
contextTokens: contextTokensUsed ?? entry.contextTokens,
|
|
updatedAt: Date.now(),
|
|
}),
|
|
});
|
|
} catch (err) {
|
|
logVerbose(
|
|
`failed to persist followup model/context update: ${String(err)}`,
|
|
);
|
|
}
|
|
}
|
|
}
|
|
|
|
await sendFollowupPayloads(finalPayloads, queued);
|
|
} finally {
|
|
typing.markRunComplete();
|
|
}
|
|
};
|
|
}
|