Files
PaperClipAI/server/src/services/github-connection-events.ts
T
DottaandPaperclip 12f70d0d4c Keep Slack form custody and receipt transfer durable (#19825)
## Thinking Path

> - Paperclip manages work for AI-agent companies.
> - Chat connections deliver provider messages and user answers.
> - Slack requires a response within its original three-second deadline.
> - Forms and their delivery receipts must survive retries and tenant
restarts.
> - This change uses the existing connector queue and durable form
owners.
> - Current authorization, exact receipt checks and normal ACK rules
stay in force.
> - The tenant can settle useful work while keeping the connection
enabled.

## Linked Issues or Issue Description

Refs #19019 and #19348. Companion Cloud PR:
paperclipai/paperclip-cloud#1250.

## What Changed

- Preserve form grants, sessions, encrypted evidence, terminal receipts
and company-authorized Retry.
- Check initial form enrollment before the provider send. Keep a typed
preparation wait only while an exact current enrollment obligation
exists.
- Refund the attempt only when the claimed attempt and credential lease
still match and provider invocation has not started. Possible or
accepted sends retain uncertain-delivery custody. An ordinary error with
the same text still uses the normal five-attempt limit.
- Notify the existing publication worker when normal enrollment commits.
Reconstruct pending work from durable state after a lost reply or owner
restart.
- Page settled history before proving idle safety. Keep unfinished,
malformed and foreign receipts visible beyond the first 1,000 rows.
- Keep Slack callback, answer, receipt and ACK rules in their provider
owners. Support body-bound Cloud idle capabilities when a complete
catalog exceeds the header budget.
- Add focused regression coverage, an enabled quiet/reactivation recipe
and explicit fixture port checks.
- Keep SDK evidence tests tied to a normally built reviewed reference,
with a separate strict refusal for actual unreviewed CI outputs.

## Verification

App head: `6284d6881b5dc38a9ee149bcbde35e69a67dd8ca`, including actual
master `dff8bed007e01c32f3826144b893ec80603d1335`. Companion Cloud head:
`f83b38187dc7e6b2e219fe865dc16d268cc66fdb`, including actual Cloud
master `b1dcc3421a91efffee631a61536cf48f357c22fc`. Cloud #1250 is merged
at `1d07d81f23d4627ab7b6dcf49176b4128a2f8f9b` after exact-head passing
checks and complete Greptile 5/5. Current exact-head CI typecheck,
build, tests and aggregate verification are green; the complete edited
Greptile review is 5/5 and all 13 review threads and nested comment
pages are resolved.

- Both latest enrollment findings were reproduced against the preceding
product. The pure readiness control returned no wait for the first
enrollment. The real-PostgreSQL control observed a provider call while
that enrollment was still staged.
- The repaired source passes three whole files: 21 readiness controls,
80 real-PostgreSQL milestone controls and 37 tenant-actor middleware
controls. All 138 pass with zero excluded or TODO cases. The SQL case
reconstructs the worker through seven waits, preserves the attempt
budget, recovers a controlled lost enrollment reply, and commits one
provider-port call and one exact grant. Replaying the selected attempt
does not duplicate them. Authorization, SDK and enrollment-response
ports are controlled in that SQL case.
- The same files cover ordinary five-attempt exhaustion, accepted and
possible sends, stale ownership, malformed bindings, attention states
and current actor restrictions. One synthetic mismatched-binding
parameter uses generation 2 against current generation 1; it is not
evidence for an older-generation control.
- The normal server-package typecheck passes at the measured source. It
runs the normal Runner TypeScript, protocol, SDK and declaration
prerequisites and all seven main App programs. The separate declaration
configuration is also recorded. A worker union-narrowing error was
found, fixed with an explicit callable/null branch, and verified by that
normal compiler. No cast or test relaxation was used.
- The normal merge of current master preserves all 78 owned paths and
15,783 unrelated master entries and modes. The three measured test files
and their changed feature paths are unchanged. New upstream DB code adds
only the normal nonunique concurrent native-session index and its
migration classification. Shared and adapter-utils changes alter
unrelated prompt or denial wording. The 138-case proof carries for those
unchanged behaviors; it is not reported as a rerun after the merge. The
normal Shared/SDK ensure, DB build, adapter-utils build and complete
server-package check also pass at the current merged head. Their fresh
inventory contains 8,143 outputs; canonical SQL and journal copies
match. This build did not apply the new index migration.
- Earlier real-PostgreSQL retained-history coverage reproduced the
missing proof with 1,005 settled rows. Its repair retains the existing
two-second/64-query inspection budget, current endpoint bound and exact
owner snapshot. The 46 source-bound controls cover unfinished and
invalid tails, exact cross-page action references, Retry, rollback and
lost receipt responses.
- Earlier enabled quiet/reactivation coverage produced ordinary useful
work, kept configuration enabled, accepted fresh signed work plus a
duplicate, and produced exactly one second useful reply. Its two
App-worker windows lasted at least 30 seconds with zero measured PG
writes, connection or outbound activity after settlement. That evidence
stays tied to its measured revision. A missing worker EXIT marker limits
that earlier proof; parent evidence does not replace it.
- Earlier three-second callback, terminal ACK, current-question race and
complete body-size controls remain source-bound. The
500-owner/1,000-binding body measured 313,875 bytes with a 618-byte JWT;
signature tampering failed closed.
- Final managed-provider acceptance now passes all five fresh Node
boots: A/A, A/AB, B/A rejection, B/AB recovery, B/B final. Two
independently configured, challenged, verified, linked and enrolled
Slack identities exercise ACK and retirement debt separately. Normal
provider factory/materialization delivers the configuration through a
controlled external stack-provider boundary. The registered opening
returns HTTP 200 in 21 ms within the original 3-second deadline. Two
answers persist, normal ACK is accepted, exact retirement and lost
acceptance/GC replies recover, and the original canonical snapshot
remains unchanged without duplicate provider sends or submissions.
- The separate complete committed-preflight case passes. A committed
`update_attempted` response is lost before SDK entry; normal
service-owner reconstruction after the original lease/deadline recovers
the same useful update and one answer without a second trigger. This
reconstructs service owners within one App process; the separate five
fresh Node boots cover process restart and managed signing-key rotation.
- Assertion-time PG socket reports balance 15 opens/15 closes across the
five boots and 5/5 in preflight, peak 2 and zero live sockets or
unexpected attempts. Worker terminal EXIT markers were missing and are
not replaced by parent EXIT markers. These are domain/recovery and
assertion-time connection proofs, not a new whole-worker terminal quiet
measurement. All known owners/groups closed; independent SQL audits were
empty; only successful fixture databases were removed. Earlier failed
fixture databases/logs remain retained. No product guard, deadline or
assertion was weakened. Exact runtime source and normal outputs are
unchanged by the three test files in the current follow-up, so these
source-bound proofs carry without repeating those journeys.
- The preceding App CI ran 11,312 support cases: 11,289 passed, three
SDK-evidence cases failed, and 20 external cases skipped. Normal CI
dependency setup resolved six stale Slack patch-hash references and
produced an unreviewed lock. The public SDK proof correctly refused that
lock; the tests incorrectly assumed the CI output was a reviewed
reference. The three-file test repair keeps the protected lock policy,
source/compiler/parser/Zod pins, scanner and redaction rules unchanged.
It builds an owned reviewed reference using the existing normal
Shared/SDK producers for the positive seal test, while a separate
negative requires actual unreviewed-build privacy refusal and no
retained SDK bytes. It does not relabel the CI build as certified or
admit a new lock hash.
- All four whole SDK support files pass locally: public type 30,
evidence 16, public-input synchronizer 8 and lock preparation 10,
totaling 64 with zero excluded/TODO cases in 28.73 seconds. The genuine
fresh reference build/positive seal and separate invalidated-reference
privacy refusal passed within their original 120-second bounds. Original
installed dependencies, SDK outputs/cache and runtime source stayed
unchanged; normal global setup produced only its own Runner JavaScript
outputs. PostgreSQL was denied throughout, all four worker markers were
observed, and the owned process group/private test root closed. The
local canonical-lock negative mutates only its own reference; fresh
remote CI also passes the actual resolved-CI-lock branch: all 16
evidence and 30 public-type tests pass, including the genuine
reviewed-reference positive and the strict actual-unreviewed-build
refusal. The upstream synchronizer file runs its current 10 cases in CI.
- The first complete CI attempt on this head also exposed two separate
30-second timeouts: one real reporter-artifact test and the
`agent-test-environment` module-loading hook. Full original logs remain
preserved. Their exact test, sanitizer and evidence source blobs are
unchanged from the preceding head; inspection found no deterministic
setup cause. After the original run finished, one retry of only the
failed jobs passed on the same synthetic merge
`168851e210671ebe97e7a778e8cf891807e80502`: Runner support reports
11,304 passes and 20 external skips; serialized shard 6 reports all 395
passes, including all 67 route bodies. No source, timeout or assertion
changed. This proves recovery; the original causes remain unexplained.
- The final pre-merge tree audit uses actual master
`17136a6c25ef6cb7c816a29140a4c48c2c75fb31` (refreshed after the earlier
`85dec046` audit). The expected normal squash preserves all 81 owned
paths and 15,816 unrelated master entries and modes. Only OpenAPI has
edits from both sides, and its complete content composition preserves
the new project route and token help alongside this PR's API entries.
The shared inbound selection changes preserve the same candidate set;
new concurrent indexes, Native/workspace behavior and shutdown changes
remain upstream behavior. The scoped form proof uses process agents and
does not qualify the new Native/sandbox branches. The latest
question-schema change still accepts the explicit version-1 payload
exercised here. The last master delta also changes managed AI/Native
credential refresh and writeback; those branches are outside the
measured process-agent form paths. Existing 8,143 local outputs remain
tied to their measured revision; no new whole-tree output equality or
deployed suspension is claimed.
- Reproduce the public path with
`doc/testing/slack-cloud-qualification.md`, the companion Cloud
checkout, Node 24/22 and fresh owned fixture databases. Aggregate local
`pnpm -r typecheck`, `pnpm test:run` and `pnpm build` were not rerun as
one command set. Current exact-head required CI and the complete
Greptile 5/5 review now pass; the full edited summaries and all
paginated threads were read.

Local worker quiet does not prove deployed host or Neon suspension.
Photon and Discord remain separate work; this PR completes the described
Slack slice.

## Risks

Use compatible App and Cloud readers before transferring custody.
Signatures prove authenticity; current authorization and exact ownership
still control execution. Preserve old targets, secrets and terminal
facts until their normal owners release them. Keep the managed seal
identity stable while custody remains. The required Slack adapter patch
remains pinned. Upstream's concurrent index uses the normal top-level
migration path.

## Model Used

OpenAI `gpt-6.1-sol` implemented and tested the changes. A GPT-6-based
Codex managing agent reviewed and integrated them. The host does not
expose the managing agent's exact model ID or context-window sizes. The
agents used code execution and test tools.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used, its version and tool capabilities
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have linked public related issues or PRs
- [x] I have not referenced internal or instance-local Paperclip issues
or links
- [x] My branch name describes the change and contains no internal
Paperclip ticket id
- [x] I have run the local tests listed above and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation
- [x] I have considered and documented risks
- [x] All Paperclip CI gates are green for the current head
- [x] Greptile is 5/5 with no open findings for the current head
- [x] I will address all reviewer comments before requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-11-10 12:17:50 -06:00

678 lines
31 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import { parseSlackFormOperation } from "./chat-channels/slack/cloud-forms.js";
import { acknowledgeSlackFormDeliveries } from "./chat-channels/slack/form-normal-ack.js";
import { readSlackFormRetainedConsumer } from "./chat-channels/slack/cloud-consumer.js";
import { reconcileRegisteredSlackFormWork } from "./chat-channels/slack/form-owner.js";
import { SlackCloudReceiptError } from "./chat-channels/slack/cloud-event.js";
import { reconcilePendingSlackCloudEnrollments } from "./chat-channels/slack/cloud-enrollment.js";
import { notifyExternalObjectRefresh } from "./external-object-refresh-notifications.js";
import { notifyGitHubCredentialChange } from "./github-connection-notifications.js";
import { notifyToolConnectionHealth } from "./tool-connection-health-notifications.js";
import { dispatchGitHubBotCloudEvent } from "./chat-github-cloud-ingress.js";
import { dispatchSlackCloudEvent } from "./chat-channels/slack/cloud-ingress.js";
import { chatEndpoints } from "@paperclipai/db";
import { randomUUID } from "node:crypto";
import { isDeepStrictEqual } from "node:util";
import {
connectionEventDeliveries,
connectionGrants,
externalObjects,
toolConnections,
type Db,
} from "@paperclipai/db";
import { and, eq, ne, sql } from "drizzle-orm";
import {
logActivity,
publishActivity,
type ActivityPublication,
} from "./activity-log.js";
import {
createPaperclipCloudConnector,
paperclipCloudConnectorConfigFromEnv,
type PaperclipCloudConnector,
type SealedConnectorEvents,
} from "./paperclip-cloud-connector.js";
import { issueThreadInteractionService } from "./issue-thread-interactions.js";
import { logger } from "../middleware/logger.js";
type LeasedEvent = SealedConnectorEvents["events"][number];
type GitHubBinding = {
id: string;
companyId: string;
connectionId: string;
grantId: string;
subject: string;
installationId: string;
providerTenant: NonNullable<typeof connectionGrants.$inferSelect.providerTenant>;
};
export type GitHubConnectionEventPollResult = {
leased: number;
processed: number;
duplicate: number;
ignored: number;
failed: number;
};
function record(value: unknown): Record<string, unknown> {
return value && typeof value === "object" && !Array.isArray(value)
? value as Record<string, unknown>
: {};
}
function stringValue(value: unknown): string | null {
return typeof value === "string" && value.length > 0 ? value : null;
}
function positiveInteger(value: unknown): number | null {
return typeof value === "number" && Number.isSafeInteger(value) && value > 0 ? value : null;
}
function boundedString(value: unknown, maximum: number): string | null {
return typeof value === "string" && value.length > 0 && value.length <= maximum ? value : null;
}
function identifier(value: unknown): string | null {
return typeof value === "string" && /^[1-9][0-9]{0,30}$/.test(value) ? value : null;
}
function isoDate(value: unknown): string | null {
const candidate = boundedString(value, 100);
if (!candidate || Number.isNaN(Date.parse(candidate))) return null;
return new Date(candidate).toISOString();
}
function commitSha(value: unknown): string | null {
return typeof value === "string" && /^[0-9a-f]{40,64}$/i.test(value) ? value.toLowerCase() : null;
}
function githubUrl(value: unknown): string | null {
const candidate = boundedString(value, 2_000);
if (!candidate) return null;
try {
const url = new URL(candidate);
return url.protocol === "https:" && url.hostname.toLowerCase() === "github.com" ? url.toString() : null;
} catch {
return null;
}
}
function compact(values: Record<string, unknown>): Record<string, unknown> {
return Object.fromEntries(Object.entries(values).filter(([, value]) => value !== null && value !== undefined));
}
/**
* Treat the sealed Cloud batch as an untrusted boundary. Cloud already normalizes
* GitHub payloads, but the instance independently allowlists and bounds the small
* reconciliation record that it persists and processes.
*/
function normalizeLeasedPayload(event: LeasedEvent): Record<string, unknown> {
const payload = record(event.payload);
const base = {
event: boundedString(payload.event, 100),
action: boundedString(payload.action, 100),
installationId: identifier(payload.installationId),
repositoryId: identifier(payload.repositoryId),
repository: boundedString(payload.repository, 300),
senderId: identifier(payload.senderId),
senderLogin: boundedString(payload.senderLogin, 100),
};
if (event.event === "pull_request") {
return compact({
...base,
number: positiveInteger(payload.number),
url: githubUrl(payload.url),
state: boundedString(payload.state, 40),
merged: payload.merged === true,
mergedAt: isoDate(payload.mergedAt),
updatedAt: isoDate(payload.updatedAt),
headRef: boundedString(payload.headRef, 300),
headSha: commitSha(payload.headSha),
baseRef: boundedString(payload.baseRef, 300),
baseSha: commitSha(payload.baseSha),
});
}
if (event.event === "installation_repositories") {
const repositoryIds = (value: unknown) => Array.isArray(value)
? value.slice(0, 1_000).flatMap((item) => identifier(item) ?? [])
: [];
return compact({
...base,
repositorySelection: boundedString(payload.repositorySelection, 40),
repositoriesAdded: repositoryIds(payload.repositoriesAdded),
repositoriesRemoved: repositoryIds(payload.repositoriesRemoved),
});
}
if (event.event === "installation") {
return compact({
...base,
accountId: identifier(payload.accountId),
accountLogin: boundedString(payload.accountLogin, 100),
repositorySelection: boundedString(payload.repositorySelection, 40),
});
}
return compact(base);
}
export function githubConnectionEventBindings(rows: Array<{
grant: typeof connectionGrants.$inferSelect;
connection: typeof toolConnections.$inferSelect;
}>): GitHubBinding[] {
return rows.flatMap(({ grant, connection }) => {
const config = record(connection.config);
const oauth = record(config.oauth);
if (config.sourceTemplateKey !== "github" || oauth.connectorProfile !== "github.code") return [];
const github = grant.providerTenant?.github;
if (!github || grant.status !== "active") return [];
const subject = grant.kind === "agent" && grant.subjectAgentId
? `agent:${grant.subjectAgentId}`
: grant.kind === "user" && grant.subjectUserId
? grant.subjectUserId
: null;
if (!subject) return [];
return github.installationIds.map((installationId) => ({
id: `${grant.id}_${installationId}`,
companyId: grant.companyId,
connectionId: connection.id,
grantId: grant.id,
subject,
installationId,
providerTenant: grant.providerTenant!,
}));
});
}
function githubSnapshotUpdate(payload: Record<string, unknown>) {
const repository = stringValue(payload.repository);
const number = positiveInteger(payload.number);
if (!repository || !number) return null;
const state = stringValue(payload.state) ?? "unknown";
const merged = payload.merged === true;
const [owner, repo, ...extra] = repository.split("/");
if (!owner || !repo || extra.length > 0) return null;
return {
repository,
owner,
repo,
number,
externalId: `${repository}#pull/${number}`,
state,
merged,
statusKey: merged ? "merged" : state === "closed" ? "closed" : "open",
statusLabel: merged ? "Merged" : state === "closed" ? "Closed" : "Open",
statusCategory: merged ? "succeeded" : state === "closed" ? "closed" : "open",
statusTone: merged ? "success" : state === "closed" ? "muted" : "info",
statusIconKey: merged ? "git-merge" : state === "closed" ? "x-circle" : "git-pull-request",
data: {
provider: "github",
owner,
repo,
number,
state,
merged,
...(stringValue(payload.url) ? { url: stringValue(payload.url) } : {}),
...(stringValue(payload.mergedAt) ? { mergedAt: stringValue(payload.mergedAt) } : {}),
...(stringValue(payload.headRef) ? { headRef: stringValue(payload.headRef) } : {}),
...(stringValue(payload.headSha) ? { headSha: stringValue(payload.headSha) } : {}),
...(stringValue(payload.baseRef) ? { baseRef: stringValue(payload.baseRef) } : {}),
...(stringValue(payload.baseSha) ? { baseSha: stringValue(payload.baseSha) } : {}),
},
remoteVersion: stringValue(payload.updatedAt),
} as const;
}
export function githubConnectionEventService(
db: Db,
options: {
connector?: PaperclipCloudConnector;
env?: NodeJS.ProcessEnv;
now?: () => Date;
wakeup?: NonNullable<Parameters<typeof issueThreadInteractionService>[1]>["wakeup"];
} = {},
) {
const now = options.now ?? (() => new Date());
let nextPollAt = 0;
let nextEnrollmentAt: number | null = null;
let nextEnrollmentDiagnosticAt = 0;
const slackDiagnostics = new Map<string, number>();
let configuredConnector: {
config: NonNullable<ReturnType<typeof paperclipCloudConnectorConfigFromEnv>>;
connector: PaperclipCloudConnector;
} | undefined;
function currentConnector() {
if (options.connector) return options.connector;
// Enrollment and identity replacement can happen after this owner starts.
// Read only on an admitted pass; preserve negotiated delivery while unchanged.
const config = paperclipCloudConnectorConfigFromEnv(options.env);
if (!config) { configuredConnector = undefined; return null; }
if (!configuredConnector || !isDeepStrictEqual(configuredConnector.config, config)) {
configuredConnector = { config, connector: createPaperclipCloudConnector({ config }) };
}
return configuredConnector.connector;
}
const forcePolling = (options.env ?? process.env).PAPERCLIP_GITHUB_EVENT_POLLING === "1";
let emptyPolls = 0;
let nextSetupPollAt = 0;
let inFlight: Promise<GitHubConnectionEventPollResult> | null = null;
async function activeBindings() {
const rows = await db.select({ grant: connectionGrants, connection: toolConnections })
.from(connectionGrants)
.innerJoin(toolConnections, and(
eq(toolConnections.id, connectionGrants.connectionId),
eq(toolConnections.companyId, connectionGrants.companyId),
))
.where(and(
sql`${toolConnections.config}->>'sourceTemplateKey' = 'github'`,
sql`${toolConnections.config}->'oauth'->>'connectorProfile' = 'github.code'`,
eq(connectionGrants.status, "active"),
eq(toolConnections.status, "active"),
eq(toolConnections.enabled, true),
sql`exists (select 1 from companies where id = ${connectionGrants.companyId} and status <> 'archived')`,
));
return githubConnectionEventBindings(rows);
}
async function applyPullRequestEvent(database: Db, companyId: string, event: LeasedEvent, effects: Array<(committedDb: Db) => Promise<void>>) {
const snapshot = githubSnapshotUpdate(event.payload);
if (!snapshot) return;
const appliedAt = now();
await notifyExternalObjectRefresh(database, companyId, { providerKey: "github", objectType: "pull_request",
externalId: snapshot.externalId, ignoreExternalIdCase: true });
await database.update(externalObjects).set({
statusKey: snapshot.statusKey,
statusLabel: snapshot.statusLabel,
statusCategory: snapshot.statusCategory,
statusTone: snapshot.statusTone,
statusIconKey: snapshot.statusIconKey,
isTerminal: snapshot.merged || snapshot.state === "closed",
data: sql`${externalObjects.data} || ${JSON.stringify(snapshot.data)}::jsonb`,
remoteVersion: snapshot.remoteVersion,
lastResolvedAt: appliedAt,
lastChangedAt: appliedAt,
nextRefreshAt: appliedAt,
refreshStartedAt: null,
refreshToken: null,
updatedAt: appliedAt,
}).where(and(
eq(externalObjects.companyId, companyId),
eq(externalObjects.providerKey, "github"),
eq(externalObjects.objectType, "pull_request"),
sql`lower(${externalObjects.externalId}) = lower(${snapshot.externalId})`,
));
if (snapshot.merged && event.action === "closed") {
await issueThreadInteractionService(database, { wakeup: options.wakeup })
.sweepMergedPullRequestConfirmations([{
companyId,
owner: snapshot.owner,
repo: snapshot.repo,
number: snapshot.number,
}], { companyId }, undefined, { deferConfirmationCommitEffects: effect => effects.push(effect) });
}
}
async function applyInstallationEvent(database: Db, binding: GitHubBinding, event: LeasedEvent) {
// Bindings are loaded before the Cloud request. Lock and read the grant
// again so a refresh completed during that request cannot be overwritten.
const [currentGrant] = await database.select().from(connectionGrants).where(and(
eq(connectionGrants.id, binding.grantId),
eq(connectionGrants.companyId, binding.companyId),
eq(connectionGrants.status, "active"),
)).for("update").limit(1);
const currentProviderTenant = currentGrant?.providerTenant;
const github = currentProviderTenant?.github;
if (!github) return;
// A newly bound instance can receive installation events from before OAuth
// verified its repository list. Those events must not erase newer access.
if (Date.parse(github.lastAccessRefreshAt ?? "") > Date.parse(event.createdAt)) {
await database.update(connectionGrants).set({
providerTenant: {
...currentProviderTenant,
github: { ...github, lastWebhookAt: now().toISOString(), webhookHealth: "healthy" },
},
updatedAt: now(),
}).where(and(eq(connectionGrants.id, binding.grantId), eq(connectionGrants.companyId, binding.companyId)));
return;
}
const unavailable = event.event === "installation" && (event.action === "deleted" || event.action === "suspend");
const installationIds = unavailable
? github.installationIds.filter((id) => id !== binding.installationId)
: [...new Set([...github.installationIds, binding.installationId])];
const added = Array.isArray(event.payload.repositoriesAdded) ? event.payload.repositoriesAdded.length : 0;
const removed = Array.isArray(event.payload.repositoriesRemoved) ? event.payload.repositoriesRemoved.length : 0;
const repositoryCount = Math.max(
0,
unavailable && installationIds.length === 0
? 0
: unavailable
? github.repositoryCount
: github.repositoryCount + added - removed,
);
const providerTenant = {
...currentProviderTenant,
github: {
...github,
// Lifecycle webhooks carry IDs, not the user token’s complete repository view.
// Discard the snapshot until Refresh access verifies it again.
accessRevision: randomUUID(),
...(unavailable || github.bindingSync || !github.installationIds.includes(binding.installationId) ? { bindingSync: {
id: randomUUID(), removeInstallationIds: [...new Set([
...(github.bindingSync?.removeInstallationIds ?? []), ...(unavailable ? [binding.installationId] : []),
])].filter(id => !installationIds.includes(id)),
} } : {}),
repositories: undefined,
installationIds,
installationCount: installationIds.length,
repositoryCount,
repositorySelection: repositoryCount === 0 ? "none" as const : github.repositorySelection,
lastWebhookAt: now().toISOString(),
webhookHealth: unavailable ? "unhealthy" as const : "healthy" as const,
},
};
await notifyGitHubCredentialChange(database, { companyId: binding.companyId, config: { sourceTemplateKey: "github" } });
await database.update(connectionGrants).set({ providerTenant, updatedAt: now() })
.where(and(eq(connectionGrants.id, binding.grantId), eq(connectionGrants.companyId, binding.companyId)));
await notifyToolConnectionHealth(database, { companyId: binding.companyId, id: binding.connectionId });
await database.update(toolConnections).set({
healthStatus: unavailable ? "failed" : "ok",
healthMessage: unavailable
? "GitHub installation access was removed or suspended. Manage repository access on GitHub."
: "GitHub installation and repository access are available.",
healthCheckedAt: now(),
lastHealthAt: now(),
lastError: unavailable ? "GitHub installation unavailable" : null,
updatedAt: now(),
}).where(and(eq(toolConnections.id, binding.connectionId), eq(toolConnections.companyId, binding.companyId)));
}
async function processForCompany(companyId: string, bindings: GitHubBinding[], event: LeasedEvent) {
const receiptAt = now();
const normalizedEvent = { ...event, payload: normalizeLeasedPayload(event) };
const [receipt] = await db.insert(connectionEventDeliveries).values({
companyId,
provider: event.provider,
providerDeliveryId: event.id,
event: event.event,
action: event.action,
installationId: event.installationId,
repositoryId: event.repositoryId,
normalizedPayload: normalizedEvent.payload,
providerCreatedAt: new Date(event.createdAt),
status: "received",
attempts: 1,
updatedAt: receiptAt,
}).onConflictDoNothing().returning();
if (!receipt) {
const [existing] = await db.select().from(connectionEventDeliveries).where(and(
eq(connectionEventDeliveries.companyId, companyId),
eq(connectionEventDeliveries.provider, event.provider),
eq(connectionEventDeliveries.providerDeliveryId, event.id),
)).limit(1);
if (existing?.status === "processed") return "duplicate" as const;
}
const postCommitPublications: ActivityPublication[] = [];
const postCommitEffects: Array<(committedDb: Db) => Promise<void>> = [];
let applied = true;
try {
const applyAndFinalize = async (database: Db) => {
const [locked] = await database.select().from(connectionEventDeliveries).where(and(
eq(connectionEventDeliveries.companyId, companyId), eq(connectionEventDeliveries.provider, event.provider),
eq(connectionEventDeliveries.providerDeliveryId, event.id),
)).for("update").limit(1);
if (locked!.status === "processed") { applied = false; return; }
if (!receipt) await database.update(connectionEventDeliveries).set({
status: "received", attempts: sql`${connectionEventDeliveries.attempts} + 1`, lastError: null, updatedAt: receiptAt,
}).where(eq(connectionEventDeliveries.id, locked!.id));
if (event.event === "pull_request") await applyPullRequestEvent(database, companyId, normalizedEvent, postCommitEffects);
if (event.event === "installation" || event.event === "installation_repositories") {
for (const binding of bindings) await applyInstallationEvent(database, binding, normalizedEvent);
} else {
const touchedAt = now();
for (const binding of bindings) {
const github = binding.providerTenant.github;
if (!github) continue;
await database.update(connectionGrants).set({
// Update only webhook fields; a concurrent access/token refresh
// owns the remaining metadata and must not be replaced here.
providerTenant: sql`jsonb_set(${connectionGrants.providerTenant}, '{github}',
(${connectionGrants.providerTenant}->'github') || ${JSON.stringify({
lastWebhookAt: touchedAt.toISOString(), webhookHealth: "healthy",
})}::jsonb)`,
updatedAt: touchedAt,
}).where(and(
eq(connectionGrants.id, binding.grantId),
eq(connectionGrants.companyId, companyId),
eq(connectionGrants.status, "active"),
sql`${connectionGrants.providerTenant}->'github' is not null`,
));
}
}
const finishedAt = now();
await database.update(connectionEventDeliveries).set({
status: "processed",
processedAt: finishedAt,
lastError: null,
updatedAt: finishedAt,
}).where(and(
eq(connectionEventDeliveries.companyId, companyId),
eq(connectionEventDeliveries.provider, event.provider),
eq(connectionEventDeliveries.providerDeliveryId, event.id),
));
await logActivity(database, {
companyId,
actorType: "system",
actorId: "system:github-webhook",
action: "tool_connection.webhook_processed",
entityType: "tool_connection",
entityId: bindings[0]!.connectionId,
details: {
provider: "github",
event: event.event,
action: event.action,
deliveryId: event.id,
installationId: event.installationId,
repositoryId: event.repositoryId,
},
}, postCommitPublications);
};
await db.transaction(async (tx) => applyAndFinalize(tx as unknown as Db));
} catch (error) {
await db.update(connectionEventDeliveries).set({
status: "failed",
lastError: error instanceof Error ? error.message.slice(0, 500) : "GitHub event processing failed",
updatedAt: now(),
}).where(and(
eq(connectionEventDeliveries.companyId, companyId),
eq(connectionEventDeliveries.provider, event.provider),
eq(connectionEventDeliveries.providerDeliveryId, event.id),
ne(connectionEventDeliveries.status, "processed"),
));
throw error;
}
// Persistence is complete at this point (and installation deltas have
// committed). A synchronous live-event subscriber must not turn that
// durable success back into a retryable receipt and replay the delta.
for (const publication of postCommitPublications) {
try {
publishActivity(publication);
} catch (error) {
logger.warn({
err: error,
companyId,
providerDeliveryId: event.id,
}, "GitHub webhook activity publication failed after commit");
}
}
for (const effect of postCommitEffects) await effect(db);
return applied ? "processed" as const : "duplicate" as const;
}
async function runPoll(): Promise<GitHubConnectionEventPollResult> {
const connector = currentConnector();
if (!connector) {
nextPollAt = Number.POSITIVE_INFINITY;
return { leased: 0, processed: 0, duplicate: 0, ignored: 0, failed: 0 };
}
try {
return await drainConnectorEvents(connector);
} finally {
try {
nextEnrollmentAt = await reconcilePendingSlackCloudEnrollments(db, now());
} catch {
const time = now().getTime();
nextEnrollmentAt = time + 5_000;
if (time >= nextEnrollmentDiagnosticAt) {
nextEnrollmentDiagnosticAt = time + 300_000;
logger.warn({ reason: "enrollment_reconcile_unavailable" }, "Slack enrollment responsibility retained");
}
}
let formsAt: number | null = null;
try { formsAt = await reconcileRegisteredSlackFormWork(db, now()); }
catch { formsAt = now().getTime() + 5_000; }
const next = Math.min(nextEnrollmentAt ?? Number.POSITIVE_INFINITY, formsAt ?? Number.POSITIVE_INFINITY);
nextEnrollmentAt = Number.isFinite(next) ? next : null;
}
}
async function drainConnectorEvents(connector: PaperclipCloudConnector): Promise<GitHubConnectionEventPollResult> {
const bindings = await activeBindings();
const githubBots = await db.select().from(chatEndpoints).where(and(eq(chatEndpoints.provider, "github"), sql`${chatEndpoints.setup}->'github'->>'cloudRegistrationId' is not null`, sql`${chatEndpoints.status} not in ('archived', 'disabled')`, sql`exists (select 1 from companies where id = ${chatEndpoints.companyId} and status <> 'archived')`));
const slackBots = await db.select().from(chatEndpoints).where(and(eq(chatEndpoints.provider, "slack"), sql`${chatEndpoints.setup}->'slackCloudEvents' is not null`, sql`${chatEndpoints.status} in ('active','verifying','attention')`, sql`exists (select 1 from companies where id = ${chatEndpoints.companyId} and status <> 'archived')`));
const bots = [...githubBots, ...slackBots];
let identity: { subject: string; companyId: string } | null = bindings[0] ?? (bots[0] ? { subject: "github-bot-events", companyId: bots[0].companyId } : null);
let hasConsumer = bindings.length > 0 || bots.length > 0;
if (!identity) {
try {
const retained = await readSlackFormRetainedConsumer(db, connector);
identity = retained.owner;
hasConsumer ||= retained.owner !== null;
const time = now().getTime();
if (retained.attention.length && time >= nextEnrollmentDiagnosticAt) {
nextEnrollmentDiagnosticAt = time + 300_000;
logger.warn({ reason: "form_retained_owner_invalid", count: retained.attention.length }, "Slack form custody requires attention");
}
} catch {
const time = now().getTime();
if (time >= nextEnrollmentDiagnosticAt) {
nextEnrollmentDiagnosticAt = time + 300_000;
logger.warn({ reason: "form_retained_owner_unavailable" }, "Slack form custody requires attention");
}
}
}
if (!identity) {
const [history] = await db.select({ grant: connectionGrants, connection: toolConnections }).from(connectionGrants)
.innerJoin(toolConnections, and(eq(toolConnections.id, connectionGrants.connectionId), eq(toolConnections.companyId, connectionGrants.companyId)))
.where(and(sql`${toolConnections.config}->>'sourceTemplateKey' = 'github'`, sql`${toolConnections.config}->'oauth'->>'connectorProfile' = 'github.code'`)).limit(1);
const grant = history?.grant;
const subject = grant?.kind === "agent" && grant.subjectAgentId ? `agent:${grant.subjectAgentId}`
: grant?.kind === "user" ? grant.subjectUserId : null;
if (grant && subject) identity = { subject, companyId: grant.companyId };
}
if (!identity) {
const [history] = await db.select().from(chatEndpoints).where(sql`(${chatEndpoints.provider} = 'github' AND ${chatEndpoints.setup}->'github'->>'cloudRegistrationId' is not null) OR (${chatEndpoints.provider} = 'slack' AND ${chatEndpoints.setup}->'slackCloudEvents' is not null)`).limit(1);
if (history) identity = { subject: "github-bot-events", companyId: history.companyId };
}
if (!identity) {
nextPollAt = Number.POSITIVE_INFINITY;
return { leased: 0, processed: 0, duplicate: 0, ignored: 0, failed: 0 };
}
const first = identity;
const lease = await connector.leaseEvents({ subject: first.subject, companyId: first.companyId, eventWakeVersion: forcePolling ? 0 : 1 });
if (!lease) {
emptyPolls += 1;
nextPollAt = !hasConsumer || (!forcePolling && connector.eventWakeVersion?.() === 1) ? Number.POSITIVE_INFINITY
: now().getTime() + Math.min(bots.length ? 15_000 : 5 * 60_000, 5_000 * (2 ** Math.min(emptyPolls, 6)));
return { leased: 0, processed: 0, duplicate: 0, ignored: 0, failed: 0 };
}
emptyPolls = 0;
nextPollAt = !hasConsumer || (!forcePolling && connector.eventWakeVersion?.() === 1) ? Number.POSITIVE_INFINITY : now().getTime() + 5_000;
const result: GitHubConnectionEventPollResult = {
leased: lease.events.length + (lease.failedEventIds?.length ?? 0),
processed: 0,
duplicate: 0,
ignored: 0,
failed: lease.failedEventIds?.length ?? 0,
};
const acknowledge: string[] = [];
const formCompanies = new Map<string, string>();
for (const event of lease.events) {
if (event.provider === "slack") {
try {
await dispatchSlackCloudEvent(db, event);
if (/^slack_form_[a-f0-9]{64}$/.test(event.id)) {
const packet = event.payload.slackFormOpen;
if (!packet || typeof packet !== "object" || Array.isArray(packet)) throw new SlackCloudReceiptError("form_ack_company_missing", "attention");
formCompanies.set(event.id, parseSlackFormOperation(Reflect.get(packet, "operation")).binding.companyId);
}
acknowledge.push(event.id); result.processed += 1;
}
catch (error) {
const code = error instanceof SlackCloudReceiptError ? error.code : "owner_or_admission_unavailable";
const disposition = error instanceof SlackCloudReceiptError ? error.disposition : "retry";
const time = now().getTime();
if (time - (slackDiagnostics.get(event.id) ?? Number.NEGATIVE_INFINITY) >= 300_000) {
if (slackDiagnostics.size >= 1000) slackDiagnostics.delete(slackDiagnostics.keys().next().value!);
slackDiagnostics.set(event.id, time);
logger.warn({ deliveryId: event.id.slice(0, 200), reason: code, disposition,
recoveryExpiresAt: Number.isFinite(Date.parse(event.createdAt)) ? new Date(Date.parse(event.createdAt) + 30 * 24 * 60 * 60_000).toISOString() : null }, "Slack connector receipt requires attention");
}
if (disposition === "invalid") { acknowledge.push(event.id); result.ignored += 1; }
else result.failed += 1;
}
continue;
}
if (event.payload.githubApp !== undefined) {
try { await dispatchGitHubBotCloudEvent(db, event); acknowledge.push(event.id); result.processed += 1; }
catch { result.failed += 1; }
continue;
}
const matched = bindings.filter((binding) => event.bindingIds.includes(binding.id));
if (matched.length === 0) {
result.ignored += 1;
acknowledge.push(event.id);
continue;
}
try {
const companies = new Map<string, GitHubBinding[]>();
for (const binding of matched) companies.set(binding.companyId, [...(companies.get(binding.companyId) ?? []), binding]);
for (const [companyId, companyBindings] of companies) {
const status = await processForCompany(companyId, companyBindings, event);
result[status] += 1;
}
acknowledge.push(event.id);
} catch {
result.failed += 1;
}
}
if (acknowledge.length > 0) {
await acknowledgeSlackFormDeliveries(db, { connector,
values: { subject: first.subject, companyId: first.companyId, leaseId: lease.leaseId, deliveryIds: acknowledge }, companies: formCompanies });
}
for (const deliveryId of lease.failedEventIds ?? []) {
logger.warn({ deliveryId, reason: "inner_event_authentication_failed" },
"GitHub connector event retained for retry");
}
return result;
}
return {
nextPollAt: () => { const next = Math.min(nextPollAt, nextEnrollmentAt ?? Number.POSITIVE_INFINITY); return Number.isFinite(next) ? next : null; },
pollOnce(input: { forSetup?: boolean; forDelivery?: boolean } = {}): Promise<GitHubConnectionEventPollResult> {
// A manager actively awaiting a dedicated App's signed ping must not
// inherit a personal connection's old empty-queue backoff. This is a
// bounded readiness probe, not an additional background polling loop.
if (inFlight) return inFlight;
const time = now().getTime();
if (!input.forDelivery && time < (input.forSetup ? nextSetupPollAt : Math.min(nextPollAt, nextEnrollmentAt ?? Number.POSITIVE_INFINITY))) {
return Promise.resolve({ leased: 0, processed: 0, duplicate: 0, ignored: 0, failed: 0 });
}
nextSetupPollAt = time + 5_000;
nextPollAt = Math.max(nextPollAt, nextSetupPollAt);
const work = runPoll().finally(() => { if (inFlight === work) inFlight = null; });
inFlight = work;
return work;
},
};
}