mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-09 16:02:36 +01:00
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Routine-driven agents run on the native Runner through ACP harnesses such as Claude > - When the agent's provider account runs out of allowance, every new wake for that agent fails the same way > - The ACP harness reported the exhausted allowance as a generic "terminal limit failure" with an `unknown` category, so Paperclip could not tell it from a context or turn limit > - Paperclip recorded a reset time only for the failing run's own startup retry; other wakes of the same agent kept dispatching, each spent about two minutes on workspace setup, and then failed > - This pull request classifies the limit, keeps its reset time, and holds the agent's new work until that reset > - The benefit is that a limited agent stops burning runs, shows a clear "waiting for usage limit reset" state with a time, and resumes its queued work after the reset ## Linked Issues or Issue Description Refs #13148 Refs #13379 **What happened?** A routine-driven Claude agent on the native Runner failed 29 times in about 15 hours. 20 runs failed with `native_provider_terminal_failed` and "ACP agent reported a terminal limit failure." The recorded provider failure had category `unknown`. 6 runs failed with `provider_session_open_rejected`. Their `session.open` result is `command_execution_failed` with classification `provider_rate_limit`, and the provider failure is HTTP 429 `rate_limit` ("would exceed your account's rate limit"). The startup retry of one such run recorded a cooldown until one hour later. A new wake for a different task of the same agent dispatched 16 minutes later, spent two minutes on workspace setup, and failed with the same 429. Separately, 4 runs failed with "Project authentication conflicts with this agent's managed AI connection". That message did not say which file or setting conflicted. The same message was also shown when the check could not finish. **What did you expect?** A provider usage limit is classified distinctly. Paperclip records the reset time when the provider reports one. New work for that agent waits for the reset with a visible reason. Queued work resumes after the reset, and no doomed run starts. A configuration conflict is reported only when one is found, and the message names the file to fix. **Steps to reproduce** 1. Use a native Runner agent with a Claude ACP harness whose account is rate limited or out of usage. 2. Let a routine or assignment wake the agent on several tasks. 3. See each run fail with a generic terminal limit failure or a `session.open` rejection, and see new runs keep starting. ## What Changed - Runner: The ACPX turn now receives the typed AIR session failure (`onTerminalSessionFailure`) through the runtime host and the ACPX port. The sidecar adds the provider's title to the provider failure message. Only a rate-limit or quota title becomes `rate_limit` or `quota_exhausted`. Context, turn and configured budget limits stay `unknown`. The failure details stay private. - Server: A native `turn.failed` with a `rate_limit` or `quota_exhausted` provider failure is now `native_provider_usage_limit`, the same code as the Codex usage-limit terminal. - Server: `agentProviderLimitCooldown` (extracted from the startup retry) also reads a reset clock in the provider's wording, such as "resets 2am (UTC)". The clock parser moved from the recovery service to `provider-quota-reset.ts` without changes. - Server: New `heartbeat/provider-limit-wait.ts`. Before any workspace or provider work, a queued run for an agent with an active provider limit is cancelled as `provider_limit_wait`. The run records the reset time, whether the provider reported it, the category and the run that reported the limit. Paperclip schedules one retry of the same wake at the reset through the existing AI connection wait lane. That lane does not consume the failure retry budget. Manual runs, recovering runs, conversation turns and pool-routed connections are not held. - Server: The execution projection shows "Waiting for usage limit reset" with the reset time and the next step. - Server: Only a found project authentication conflict is a configuration failure. Its message names the project file and the settings to remove. A check that cannot finish (timeout or another exit code) is now an ordinary failure, not a configuration conflict. - Docs: `doc/execution-semantics.md` describes the agent-level limit wait. ## Verification Red before the fix, green after: - `server/src/__tests__/heartbeat-provider-limit-wait.test.ts`: An automation wake after a 429 `session.open` rejection dispatched the adapter before the fix. Now the run is cancelled as `provider_limit_wait` and a scheduled retry exists at the reset. After the reset passes, the retry is promoted and the work succeeds. A manual run, a later success, expired evidence and an unclassified terminal still dispatch. - `native-startup-retry.integration.test.ts`: "reads the reset clock from an ACP terminal limit failure's provider wording" (it fell back to one hour before). - `native-session-executor.test.ts`: An ACP `turn.failed` with a typed `quota_exhausted` or `rate_limit` failure is a usage-limit terminal; `unknown` is not. - `packages/paperclip-runner/src/drivers/provider-failure.test.ts`: A typed AIR limit title is kept and classified; context, turn and budget limits are not; details do not leak. - `execution-projection.test.ts`, `ai-connections.test.ts`: The wait label, and conflict versus unfinished check. - Safety cases: `runtime-host.test.ts` and `codex-runtime-adapter.test.ts` forward the callback. The existing startup retry, provider capacity, recovery classification, legacy reconciliation, retry scheduling and execution configuration suites pass. Commands: - `cd server && npx vitest run src/__tests__/heartbeat-provider-limit-wait.test.ts src/__tests__/ai-connections.test.ts src/services/native-runtime/ src/services/recovery/ src/services/execution-projection.test.ts src/services/legacy-execution-recovery.test.ts src/__tests__/heartbeat-retry-scheduling.test.ts src/__tests__/heartbeat-execution-configuration.test.ts` - `cd packages/paperclip-runner && npx vitest run src/drivers/provider-failure.test.ts src/drivers/acpx/runtime-host.test.ts src/drivers/acpx/codex-runtime-adapter.test.ts src/cli/acpx-runtime-sidecar.test.ts` - `tsc --noEmit` for server and paperclip-runner; the ACPX sidecar and provider-failure contract checks pass. I did not run a live provider limit on staging. The live evidence is the staging run records for the affected agent, which I read only. ## Risks - The limit evidence is per agent, not per account. If the board changes the agent's AI connection, automatic wakes can still wait until the old reset (at most one hour when no reset was reported, and at most one day). A manual run is not held, and one successful run clears the evidence. - A new wake that arrives during a limit is now cancelled and retried at the reset, instead of failing. The task stays in its status, and the run shows the reset time. - The Runner change takes effect when the provider pack is rebuilt and deployed. Until then, the server still holds new work using the existing `rate_limit` evidence from `session.open` rejections. - A turn terminal with a typed rate limit or quota is now `native_provider_usage_limit` instead of `native_provider_terminal_failed`. Both are terminal and board-owned. The recovery text now asks the board to restore provider usage capacity. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, used through Claude Code with tool use (shell, file edits, browser reads of staging run records) and extended reasoning. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [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 either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
667 lines
32 KiB
TypeScript
667 lines
32 KiB
TypeScript
import { environmentTaskResultSchema } from "@paperclipai/plugin-sdk";
|
|
import type { PluginTaskCredentialWritebackBinding } from "./plugin-task-credential-writeback.js";
|
|
import { OPENROUTER_ATTRIBUTION_HEADERS, openRouterClaudeHeaders } from "@paperclipai/shared/openrouter-attribution";
|
|
import { createHash } from "node:crypto";
|
|
import { HttpError, unprocessable } from "../errors.js";
|
|
import { mkdtemp, mkdir, writeFile, readFile, realpath, rm } from "node:fs/promises";
|
|
import os from "node:os";
|
|
import path from "node:path";
|
|
import { and, eq } from "drizzle-orm";
|
|
import { type Db, companySecrets, connectionGrants } from "@paperclipai/db";
|
|
import {
|
|
AI_CONNECTION_CAPABILITIES,
|
|
aiConnectionMetadataSchema, aiRoutingHarness,
|
|
type AiConnectionBinding,
|
|
} from "@paperclipai/shared";
|
|
import { managedProviderRouting } from "./ai-provider-routing.js";
|
|
import { aiConnectionService } from "./ai-connections.js";
|
|
import { subscriptionService } from "./subscriptions.js";
|
|
import { refreshSubscriptionConnection } from "./subscription-refresh.js";
|
|
import { secretService } from "./secrets.js";
|
|
import { decideCodexAuthMerge, withAccountHomeSecretMutationLock } from "@paperclipai/adapter-codex-local/server";
|
|
import { WORKSPACE_RESTORE_LOCK_TIMEOUT_CODE } from "@paperclipai/adapter-utils/workspace-restore-merge";
|
|
import type { AdapterExecutionTarget } from "@paperclipai/adapter-utils/execution-target";
|
|
import { runAdapterExecutionTargetProcess } from "@paperclipai/adapter-utils/execution-target";
|
|
import { decideGrokAuthMerge } from "@paperclipai/adapter-grok-local/server";
|
|
import { NativeEnvironmentProbeCleanupError, sanitizeAntigravityCredential } from "../vendor/paperclip-runner/index.js";
|
|
import { readLocalAiCredentialFile } from "./local-ai-credential-file.js";
|
|
import { prepareAntigravityCredentialWriteback } from "./native-runtime/antigravity-runner-credentials.js";
|
|
import { quotaCredentialHash, quotaCredentialRecovery, RetiredCredentialRecoveryError } from "./quota-credential-recovery.js";
|
|
import { bindRetainedAiCredentialHome, reconcileRetainedAiCleanup, retainAiCredentialCleanup } from "./retained-ai-credential-cleanup.js";
|
|
|
|
export function isAiConnectionBusy(error: unknown): error is HttpError {
|
|
return error instanceof HttpError && error.status === 422 &&
|
|
(error.details as { code?: unknown } | undefined)?.code === "ai_connection_busy";
|
|
}
|
|
|
|
/** A real conflict between the agent's AI connection and its project or arguments. */
|
|
export function isAiConnectionIncompatible(error: unknown): error is HttpError {
|
|
return error instanceof HttpError && error.status === 422 &&
|
|
(error.details as { code?: unknown } | undefined)?.code === "ai_connection_incompatible";
|
|
}
|
|
|
|
function projectAuthConflict(files: string[]) {
|
|
return unprocessable(
|
|
`This agent's managed AI connection conflicts with project authentication settings in ${files.join(" or ")}. Remove the API key, token or apiKeyHelper settings from that project file, or choose a different AI connection for this agent.`,
|
|
{ code: "ai_connection_incompatible" },
|
|
);
|
|
}
|
|
|
|
// Blank values intentionally override inherited credentials in CLI child environments.
|
|
export const AI_AUTH_ENV_KEYS = [
|
|
"COPILOT_GITHUB_TOKEN",
|
|
"PAPERCLIP_AI_PROVIDER_KEY", "PAPERCLIP_AI_PROVIDER_URL", "PAPERCLIP_CODEX_PROVIDERS",
|
|
"GEMINI_API_KEY", "GOOGLE_API_KEY", "GOOGLE_GEMINI_BASE_URL", "GOOGLE_GENAI_USE_VERTEXAI",
|
|
"GEMINI_HOME", "GOOGLE_APPLICATION_CREDENTIALS", "GOOGLE_CLOUD_ACCESS_TOKEN", "PAPERCLIP_ACPX_ANTIGRAVITY_AUTH_JSON_SECRET",
|
|
"HERMES_HOME", "AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY", "AWS_SESSION_TOKEN", "AWS_BEARER_TOKEN_BEDROCK", "AWS_PROFILE", "AWS_DEFAULT_PROFILE", "AWS_SHARED_CREDENTIALS_FILE", "AWS_CONFIG_FILE", "AWS_WEB_IDENTITY_TOKEN_FILE", "AWS_ROLE_ARN", "AWS_CONTAINER_CREDENTIALS_RELATIVE_URI", "AWS_CONTAINER_CREDENTIALS_FULL_URI",
|
|
"ANTHROPIC_BEDROCK_BASE_URL", "ANTHROPIC_MODEL", "ANTHROPIC_DEFAULT_OPUS_MODEL", "ANTHROPIC_DEFAULT_SONNET_MODEL", "ANTHROPIC_DEFAULT_HAIKU_MODEL", "CLAUDE_CODE_SUBAGENT_MODEL",
|
|
|
|
"ANTHROPIC_API_KEY",
|
|
"ANTHROPIC_AUTH_TOKEN",
|
|
"CLAUDE_CODE_OAUTH_TOKEN",
|
|
"OPENAI_API_KEY",
|
|
"CODEX_API_KEY",
|
|
"OPENROUTER_API_KEY",
|
|
"XAI_API_KEY",
|
|
"GROK_API_KEY",
|
|
"CODEX_HOME",
|
|
"GROK_HOME",
|
|
"CLAUDE_CONFIG_DIR",
|
|
"OPENCODE_AUTH_JSON",
|
|
"OPENCODE_CONFIG_CONTENT",
|
|
"OPENCODE_CONFIG",
|
|
"OPENCODE_CONFIG_DIR",
|
|
"PAPERCLIP_OPENCODE_PROVIDERS",
|
|
"PAPERCLIP_PI_PROVIDERS",
|
|
"ANTHROPIC_BASE_URL",
|
|
"OPENAI_BASE_URL",
|
|
"XAI_BASE_URL",
|
|
"CLAUDE_CODE_USE_BEDROCK",
|
|
"CLAUDE_CODE_USE_VERTEX",
|
|
"CLAUDE_CODE_USE_FOUNDRY",
|
|
] as const;
|
|
export function stripAiAuthBindings(env: unknown): Record<string, unknown> {
|
|
const result = {
|
|
...(env && typeof env === "object" ? (env as Record<string, unknown>) : {}),
|
|
};
|
|
for (const key of AI_AUTH_ENV_KEYS)
|
|
if (
|
|
![
|
|
"ANTHROPIC_BASE_URL",
|
|
"OPENAI_BASE_URL",
|
|
"XAI_BASE_URL",
|
|
"CLAUDE_CODE_USE_BEDROCK",
|
|
"CLAUDE_CODE_USE_VERTEX",
|
|
"CLAUDE_CODE_USE_FOUNDRY",
|
|
"PAPERCLIP_OPENCODE_PROVIDERS",
|
|
].includes(key)
|
|
)
|
|
delete result[key];
|
|
return result;
|
|
}
|
|
export async function assertManagedAiProjectAuth(
|
|
config: Record<string, unknown>,
|
|
provider: AiConnectionBinding["provider"],
|
|
target?: AdapterExecutionTarget | null,
|
|
) {
|
|
const extraArgs = [
|
|
...(Array.isArray(config.extraArgs) ? config.extraArgs : []),
|
|
...(Array.isArray(config.args) ? config.args : []),
|
|
];
|
|
if (
|
|
extraArgs.some(
|
|
(arg) =>
|
|
typeof arg === "string" &&
|
|
/^(--config|-c|--settings|--setting-sources|--api-key|--auth-token)(=|$)/.test(
|
|
arg,
|
|
),
|
|
)
|
|
) {
|
|
throw unprocessable(
|
|
"Remove authentication/configuration overrides before selecting a managed AI connection",
|
|
{ code: "ai_connection_incompatible" },
|
|
);
|
|
}
|
|
const files =
|
|
provider === "anthropic"
|
|
? [".claude/settings.json", ".claude/settings.local.json"]
|
|
: provider === "openai"
|
|
? [".codex/config.toml"]
|
|
: [];
|
|
const pattern =
|
|
"apiKeyHelper|ANTHROPIC_API_KEY|ANTHROPIC_AUTH_TOKEN|CLAUDE_CODE_OAUTH_TOKEN|OPENAI_API_KEY|model_provider[[:space:]]*=|env_key[[:space:]]*=|experimental_bearer_token|cli_auth_credentials_store";
|
|
if (target?.kind === "remote" && target.transport === "sandbox" && target.runnerTask && files.length) {
|
|
const description = environmentTaskResultSchema.parse(await target.runnerTask({ kind: "describe" }));
|
|
if (description.kind !== "description" || description.projectAuthCheckVersion !== 1) {
|
|
throw unprocessable("The plugin environment cannot validate managed project authentication", { code: "ai_connection_incompatible" });
|
|
}
|
|
// The process owner checks the exact prepared configuration before it can submit credentials.
|
|
return;
|
|
}
|
|
if (target?.kind === "remote" && files.length) {
|
|
// Only inspect for conflicting keys; never return configuration or credential values.
|
|
const result = await runAdapterExecutionTargetProcess(
|
|
`ai-auth-check-${Date.now()}`,
|
|
target,
|
|
"sh",
|
|
[
|
|
"-c",
|
|
`
|
|
directory=$1; pattern=$2; shift 2
|
|
while :; do
|
|
for relative in "$@"; do
|
|
file="$directory/$relative"
|
|
if test -f "$file"; then
|
|
grep -Eq "$pattern" "$file"
|
|
result=$?
|
|
if test "$result" -eq 0; then exit 42; fi
|
|
if test "$result" -ne 1; then exit 43; fi
|
|
fi
|
|
done
|
|
parent=$(dirname "$directory")
|
|
if test "$parent" = "$directory"; then break; fi
|
|
directory=$parent
|
|
done`,
|
|
"ai-auth-check",
|
|
target.remoteCwd,
|
|
pattern,
|
|
...files,
|
|
],
|
|
{
|
|
cwd: target.remoteCwd,
|
|
env: {},
|
|
timeoutSec: 15,
|
|
graceSec: 1,
|
|
onLog: async () => {},
|
|
},
|
|
);
|
|
if (result.exitCode === 42) throw projectAuthConflict(files);
|
|
// The check fails closed, but an unfinished check is not a found conflict.
|
|
if (result.exitCode !== 0)
|
|
throw new Error(
|
|
`Paperclip could not check the environment's project authentication settings (${result.timedOut ? "timed out" : `exit ${result.exitCode ?? result.signal ?? "unknown"}`}). The run did not start; it can be retried.`,
|
|
);
|
|
return;
|
|
}
|
|
let directory =
|
|
typeof config.cwd === "string" ? path.resolve(config.cwd) : process.cwd();
|
|
for (;;) {
|
|
for (const relative of files) {
|
|
try {
|
|
const content = await readFile(path.join(directory, relative), "utf8");
|
|
if (
|
|
/apiKeyHelper|ANTHROPIC_API_KEY|ANTHROPIC_AUTH_TOKEN|CLAUDE_CODE_OAUTH_TOKEN|OPENAI_API_KEY|model_provider\s*=|env_key\s*=|experimental_bearer_token|cli_auth_credentials_store/.test(
|
|
content,
|
|
)
|
|
) {
|
|
throw projectAuthConflict([relative]);
|
|
}
|
|
} catch (error) {
|
|
if ((error as NodeJS.ErrnoException).code !== "ENOENT") throw error;
|
|
}
|
|
}
|
|
const parent = path.dirname(directory);
|
|
if (parent === directory) break;
|
|
directory = parent;
|
|
}
|
|
}
|
|
|
|
function managedAiHomeEnvironment(home: string): Record<string, string> {
|
|
const providerHome = path.join(home, "provider");
|
|
return {
|
|
HOME: home,
|
|
XDG_CONFIG_HOME: path.join(home, "config"),
|
|
XDG_DATA_HOME: path.join(home, "data"),
|
|
CODEX_HOME: providerHome,
|
|
GROK_HOME: providerHome,
|
|
CLAUDE_CONFIG_DIR: providerHome,
|
|
HERMES_HOME: providerHome,
|
|
};
|
|
}
|
|
|
|
/** Only the server-created credential home is volatile; retain all other config. */
|
|
export function managedAiSessionFingerprintConfig(
|
|
config: Record<string, unknown>,
|
|
managedHome: string | undefined,
|
|
): Record<string, unknown> {
|
|
if (!managedHome) return config;
|
|
const env = { ...(config.env as Record<string, unknown> | undefined) };
|
|
const stable = managedAiHomeEnvironment("<managed-ai-home>");
|
|
for (const [key, value] of Object.entries(managedAiHomeEnvironment(managedHome))) {
|
|
if (env[key] === value) env[key] = stable[key];
|
|
}
|
|
if (env.GEMINI_HOME === path.join(managedHome, "provider")) env.GEMINI_HOME = path.join("<managed-ai-home>", "provider");
|
|
const managed = config.managedAiConnection as Record<string, unknown> | undefined;
|
|
if (managed?.sessionIdentity) {
|
|
for (const key of ["ANTHROPIC_API_KEY", "ANTHROPIC_AUTH_TOKEN", "CLAUDE_CODE_OAUTH_TOKEN", "OPENAI_API_KEY", "CODEX_API_KEY", "OPENROUTER_API_KEY", "XAI_API_KEY", "GROK_API_KEY", "GEMINI_API_KEY", "GOOGLE_API_KEY", "OPENCODE_AUTH_JSON", "OPENCODE_CONFIG_CONTENT"]) {
|
|
if (env[key]) env[key] = "<managed-ai-credential>";
|
|
}
|
|
}
|
|
return { ...config, env, ...(managed?.sessionIdentity ? { managedAiConnection: { ...managed, identity: managed.sessionIdentity } } : {}) };
|
|
}
|
|
|
|
export async function prepareManagedAiRuntime(
|
|
db: Db,
|
|
input: {
|
|
companyId: string;
|
|
agentId: string;
|
|
responsibleUserId: string | null;
|
|
adapterType: string;
|
|
binding: AiConnectionBinding;
|
|
allowUninstalledPersonal?: boolean;
|
|
allowUninstalledShared?: boolean;
|
|
allowLegacyValidation?: boolean;
|
|
config: Record<string, unknown>;
|
|
},
|
|
) {
|
|
const configuredEnv =
|
|
input.config.env && typeof input.config.env === "object"
|
|
? (input.config.env as Record<string, unknown>)
|
|
: {};
|
|
for (const key of [
|
|
"ANTHROPIC_BASE_URL",
|
|
"OPENAI_BASE_URL",
|
|
"XAI_BASE_URL",
|
|
"CLAUDE_CODE_USE_BEDROCK",
|
|
"CLAUDE_CODE_USE_VERTEX",
|
|
"CLAUDE_CODE_USE_FOUNDRY",
|
|
"PAPERCLIP_OPENCODE_PROVIDERS",
|
|
"PAPERCLIP_AI_PROVIDER_URL", "PAPERCLIP_CODEX_PROVIDERS", "PAPERCLIP_PI_PROVIDERS", "OPENCODE_CONFIG_CONTENT", "OPENCODE_CONFIG", "OPENCODE_CONFIG_DIR", "GOOGLE_GEMINI_BASE_URL", "ANTHROPIC_BEDROCK_BASE_URL",
|
|
]) {
|
|
if (configuredEnv[key])
|
|
throw unprocessable(
|
|
"The configured provider routing is incompatible with this AI connection",
|
|
{ code: "ai_connection_incompatible" },
|
|
);
|
|
}
|
|
const harness = aiRoutingHarness(input.adapterType, input.config.provider, input.config.acpxAgent);
|
|
await assertManagedAiProjectAuth(input.config, harness === "claude_local" ? "anthropic" : harness === "codex_local" ? "openai" : input.binding.provider);
|
|
const service = aiConnectionService(db);
|
|
let selection = await service.select({
|
|
...input,
|
|
userId: input.responsibleUserId,
|
|
model: input.config.model,
|
|
runnerProvider: input.config.provider,
|
|
acpxAgent: input.config.acpxAgent,
|
|
});
|
|
const retainedCleanupScope = { companyId: input.companyId, accountKey: selection.grant.id };
|
|
void reconcileRetainedAiCleanup(retainedCleanupScope).catch(() => undefined);
|
|
const subscriptionFile =
|
|
selection.attribution.method === "subscription" &&
|
|
input.binding.provider !== "anthropic";
|
|
let home: string | undefined;
|
|
let releaseCredentialHome: (() => void) | undefined;
|
|
try {
|
|
const selectedGrantId = selection.grant.id;
|
|
selection = await service.select({
|
|
...input,
|
|
userId: input.responsibleUserId,
|
|
model: input.config.model,
|
|
runnerProvider: input.config.provider,
|
|
acpxAgent: input.config.acpxAgent,
|
|
});
|
|
if (selection.grant.id !== selectedGrantId)
|
|
throw unprocessable(
|
|
"The selected default changed. Retry this execution.",
|
|
);
|
|
const credentialRef = selection.grant.credentialSecretRefs.find((ref) => ref.configPath === "ai.credential");
|
|
const routing = aiConnectionMetadataSchema.parse(selection.connection.config.ai).routing;
|
|
const noAuth = routing?.auth === "none";
|
|
if (!noAuth && !credentialRef) throw unprocessable("The selected AI credential is unavailable");
|
|
const readFreshness = async () => {
|
|
if (!credentialRef) return undefined;
|
|
const [currentGrant] = await db.select({ refs: connectionGrants.credentialSecretRefs, status: connectionGrants.status })
|
|
.from(connectionGrants).where(and(eq(connectionGrants.companyId, input.companyId), eq(connectionGrants.id, selection.grant.id)));
|
|
const currentRef = currentGrant?.refs.find(ref => ref.configPath === "ai.credential");
|
|
if (currentGrant?.status !== "active" || currentRef?.secretId !== credentialRef.secretId || currentRef.versionSelector !== credentialRef.versionSelector) return undefined;
|
|
return (await db.select({ epoch: companySecrets.aiSessionEpoch, version: companySecrets.latestVersion })
|
|
.from(companySecrets).where(and(eq(companySecrets.companyId, input.companyId), eq(companySecrets.id, credentialRef.secretId))).limit(1))[0];
|
|
};
|
|
const { value, freshness } = await (async () => {
|
|
if (noAuth) return { value: "", freshness: undefined };
|
|
// Recovering a rotated quota token can advance the secret version during
|
|
// the first read. Re-read the saved credential, retaining the epoch guard
|
|
// against a concurrent reconnect or explicit rotation.
|
|
for (let attempt = 0; attempt < 2; attempt += 1) {
|
|
const before = await readFreshness();
|
|
const value = await service.runtimeCredential(selection);
|
|
const after = await readFreshness();
|
|
if (before && after && before.version === after.version && before.epoch === after.epoch) {
|
|
return { value, freshness: after };
|
|
}
|
|
if (!before || before.epoch !== after?.epoch) break;
|
|
}
|
|
throw unprocessable("The AI credential changed during preparation; retry this execution");
|
|
})();
|
|
const antigravitySubscription = harness === "antigravity_runtime" && selection.attribution.method === "subscription";
|
|
const originalCredential = antigravitySubscription ? sanitizeAntigravityCredential(value) : value;
|
|
home = await mkdtemp(
|
|
path.join(
|
|
os.tmpdir(),
|
|
`paperclip-ai-${input.companyId}-${selection.grant.id}-`,
|
|
),
|
|
);
|
|
if (harness === "antigravity_runtime") home = await realpath(home);
|
|
const providerHome = path.join(home, "provider");
|
|
await mkdir(providerHome, { mode: 0o700 });
|
|
if (retainedCleanupScope) releaseCredentialHome = bindRetainedAiCredentialHome(providerHome, retainedCleanupScope);
|
|
const env: Record<string, unknown> = {
|
|
...stripAiAuthBindings(input.config.env),
|
|
...Object.fromEntries(AI_AUTH_ENV_KEYS.map((key) => [key, ""])),
|
|
...managedAiHomeEnvironment(home),
|
|
...(input.binding.provider === "github" ? { GH_TOKEN: "", GITHUB_TOKEN: "" } : {}),
|
|
};
|
|
const capability =
|
|
AI_CONNECTION_CAPABILITIES[input.binding.provider].methods[
|
|
selection.attribution.method
|
|
]!;
|
|
const authFile = antigravitySubscription
|
|
? path.join(providerHome, "antigravity-acp", "acp_token.json")
|
|
: path.join(providerHome, "auth.json");
|
|
if (harness === "codex_local")
|
|
await writeFile(
|
|
path.join(providerHome, "config.toml"),
|
|
'cli_auth_credentials_store = "file"\n',
|
|
{ mode: 0o600 },
|
|
);
|
|
if (!routing && subscriptionFile && !antigravitySubscription) await writeFile(authFile, value, { mode: 0o600 });
|
|
else if (!routing && !subscriptionFile) env[capability.envKey] = value;
|
|
if (
|
|
!routing && input.binding.provider === "openai" && harness === "codex_local" &&
|
|
selection.attribution.method === "api_key"
|
|
) {
|
|
env.CODEX_API_KEY = value;
|
|
await writeFile(authFile, JSON.stringify({ OPENAI_API_KEY: value }), {
|
|
mode: 0o600,
|
|
});
|
|
}
|
|
if (!routing && input.binding.provider === "openrouter" && harness === "opencode_local") {
|
|
env.OPENCODE_CONFIG_CONTENT = JSON.stringify({
|
|
provider: { openrouter: { options: { apiKey: value, headers: OPENROUTER_ATTRIBUTION_HEADERS } } },
|
|
});
|
|
env.OPENCODE_DISABLE_PROJECT_CONFIG = "true";
|
|
}
|
|
if (!routing && input.binding.provider === "google" && harness === "gemini_local") {
|
|
// Gemini headless CLI requires an explicit auth choice even with an API key.
|
|
// Seed only the disposable connection home, never the operator's settings.
|
|
const geminiHome = path.join(home, ".gemini");
|
|
await mkdir(geminiHome, { mode: 0o700 });
|
|
await writeFile(path.join(geminiHome, "settings.json"), JSON.stringify({
|
|
selectedAuthType: "gemini-api-key",
|
|
security: { auth: { selectedType: "gemini-api-key" } },
|
|
}), { mode: 0o600 });
|
|
}
|
|
if (harness === "antigravity_runtime") {
|
|
// Native Antigravity reads this explicit auth selection beneath GEMINI_HOME.
|
|
// The selected AI connection owns this disposable home, not a browser login.
|
|
env.GEMINI_HOME = providerHome;
|
|
const antigravityHome = path.join(providerHome, "antigravity-acp");
|
|
await mkdir(antigravityHome, { mode: 0o700 });
|
|
await writeFile(path.join(antigravityHome, "settings.json"), JSON.stringify({
|
|
auth: { type: antigravitySubscription ? "oauth-personal" : "gemini-api-key" },
|
|
}), { mode: 0o600 });
|
|
if (antigravitySubscription) await writeFile(authFile, originalCredential, { mode: 0o600 });
|
|
}
|
|
const projected = routing ? managedProviderRouting(routing, harness, value, typeof input.config.model === "string" ? input.config.model : "") : undefined;
|
|
if (projected) {
|
|
const customHeaders = env.ANTHROPIC_CUSTOM_HEADERS ?? process.env.ANTHROPIC_CUSTOM_HEADERS;
|
|
Object.assign(env, projected.env);
|
|
if (projected.env.ANTHROPIC_CUSTOM_HEADERS && customHeaders !== undefined) {
|
|
// Preserve secret references until adapter resolution; the adapter merges
|
|
// attribution after resolving them. Ordinary gateway headers survive too.
|
|
env.ANTHROPIC_CUSTOM_HEADERS = typeof customHeaders === "string"
|
|
? openRouterClaudeHeaders(customHeaders)
|
|
: customHeaders;
|
|
}
|
|
if (projected.hermesConfig) await writeFile(path.join(providerHome, "config.yaml"), projected.hermesConfig, { mode: 0o600 });
|
|
if (projected.codexConfig) await writeFile(path.join(providerHome, "config.toml"), 'cli_auth_credentials_store = "file"\n' + projected.codexConfig, { mode: 0o600 });
|
|
}
|
|
const attribution = { ...selection.attribution } as typeof selection.attribution & { subscriptionId?: string };
|
|
if (attribution.method === "subscription") {
|
|
const subscription = {
|
|
companyId: input.companyId, connectionId: selection.connection.id, grantId: selection.grant.id,
|
|
provider: selection.attribution.provider, name: selection.connection.name,
|
|
ownerUserId: selection.grant.subjectUserId, credential: value,
|
|
session: credentialRef && freshness ? { secretId: credentialRef.secretId, epoch: freshness.epoch } : null,
|
|
};
|
|
attribution.subscriptionId = await subscriptionService(db).register(subscription);
|
|
void refreshSubscriptionConnection(db, subscription, attribution.subscriptionId);
|
|
}
|
|
const generation = createHash("sha256")
|
|
.update(originalCredential)
|
|
.digest("hex")
|
|
.slice(0, 16);
|
|
const identity = `${selection.grant.id}:${input.responsibleUserId ?? "shared"}:${generation}`;
|
|
const sessionIdentity = `${selection.grant.id}:${input.responsibleUserId ?? "shared"}:${noAuth ? "no-auth" : `${credentialRef!.secretId}:${freshness!.epoch}`}`;
|
|
let homeRemoved = false;
|
|
let ownsGoogleRecovery = false;
|
|
let removalOnly = false;
|
|
let deferredWriteback: { cause: unknown } | undefined;
|
|
const removeHome = async () => {
|
|
if (home) await rm(home, { recursive: true, force: true });
|
|
homeRemoved = true;
|
|
releaseCredentialHome?.();
|
|
};
|
|
let cleanupRunning: Promise<void> | undefined;
|
|
const cleanupRuntime = (): Promise<void> => cleanupRunning ??= (async () => {
|
|
if (homeRemoved) return;
|
|
const cleanup = async () => {
|
|
if (removalOnly) {
|
|
await removeHome();
|
|
return;
|
|
}
|
|
let googleRecovery: ReturnType<typeof quotaCredentialRecovery> | undefined;
|
|
let durableGoogleRefresh = false;
|
|
try {
|
|
if (subscriptionFile) {
|
|
const incoming = antigravitySubscription
|
|
? sanitizeAntigravityCredential(await readLocalAiCredentialFile(authFile)) : undefined;
|
|
if (antigravitySubscription && incoming !== originalCredential) {
|
|
const accountEmail = selection.grant.providerTenant?.name;
|
|
if (!credentialRef || !freshness || !accountEmail) throw new Error("The selected Google credential recovery authority is unavailable");
|
|
googleRecovery = quotaCredentialRecovery(input.companyId, selection.grant.id);
|
|
const prior = await googleRecovery.read();
|
|
if (prior && prior.value !== incoming && !ownsGoogleRecovery) {
|
|
// A previous retired owner keeps its first recoverable refresh.
|
|
// Settle it before this owner can stage another generation.
|
|
await aiConnectionService(db).runtimeCredential(selection);
|
|
}
|
|
// Physical retirement is the caller's cleanup precondition. Save
|
|
// before network verification or a database write can fail; only
|
|
// this exact grant generation may later consume the encrypted copy.
|
|
await googleRecovery.write({ companyId: input.companyId, grantId: selection.grant.id,
|
|
connectionId: selection.connection.id, secretId: credentialRef.secretId,
|
|
baseHash: quotaCredentialHash(value), value: incoming!, provider: "google",
|
|
google: { sessionEpoch: freshness.epoch, baseVersion: freshness.version,
|
|
versionSelector: credentialRef.versionSelector ?? "latest", accountEmail, verified: false } });
|
|
ownsGoogleRecovery = true;
|
|
}
|
|
// Verification itself can consume a rotating refresh. Until its
|
|
// newest result is durably journaled, the private home owns it.
|
|
durableGoogleRefresh = false;
|
|
const refreshed = antigravitySubscription
|
|
? await prepareAntigravityCredentialWriteback({ originalCredential,
|
|
refreshedCredential: incoming!, managedHome: providerHome,
|
|
accountEmail: selection.grant.providerTenant?.name,
|
|
})
|
|
: await readFile(authFile, "utf8");
|
|
if (googleRecovery) {
|
|
const pending = await googleRecovery.read();
|
|
if (!pending?.google) throw new Error("The Google credential recovery journal is unavailable");
|
|
await googleRecovery.write({ ...pending, value: refreshed, google: { ...pending.google, verified: true } });
|
|
durableGoogleRefresh = true;
|
|
}
|
|
if (refreshed !== originalCredential) {
|
|
// A quota exchange can hold this company lock for 60 seconds.
|
|
// Retry its 30-second acquisition timeout without discarding the
|
|
// provider's only copy of a rotated, single-use refresh token.
|
|
const writeBack = () => withAccountHomeSecretMutationLock(undefined, input.companyId, () => db.transaction(async (tx) => {
|
|
const [grant] = await tx
|
|
.select()
|
|
.from(connectionGrants)
|
|
.where(
|
|
and(
|
|
eq(connectionGrants.id, selection.grant.id),
|
|
eq(connectionGrants.companyId, input.companyId),
|
|
),
|
|
)
|
|
.for("update");
|
|
// A missing or revoked grant blocks the write-back. Among
|
|
// active copies, the merge decision below keeps the
|
|
// credential with the newest provider freshness field.
|
|
if (!grant || grant.status !== "active") return;
|
|
const ref = grant.credentialSecretRefs.find(
|
|
(r) => r.configPath === "ai.credential",
|
|
);
|
|
if (!ref) return;
|
|
if (antigravitySubscription && (!credentialRef || ref.secretId !== credentialRef.secretId
|
|
|| ref.versionSelector !== credentialRef.versionSelector
|
|
|| grant.providerTenant?.name?.toLowerCase() !== selection.grant.providerTenant?.name?.toLowerCase())) return;
|
|
// Lock the referenced secret row for the rest of this
|
|
// transaction. The grant-row lock above does not cover it,
|
|
// so an authorized rotation of this secret could otherwise
|
|
// land between the read and the write below and be
|
|
// overwritten by this stale write-back.
|
|
const [currentSecret] = await tx
|
|
.select({ id: companySecrets.id, epoch: companySecrets.aiSessionEpoch, version: companySecrets.latestVersion })
|
|
.from(companySecrets)
|
|
.where(
|
|
and(
|
|
eq(companySecrets.id, ref.secretId),
|
|
eq(companySecrets.companyId, input.companyId),
|
|
),
|
|
)
|
|
.for("update");
|
|
if (antigravitySubscription && (!freshness || !currentSecret || currentSecret.epoch !== freshness.epoch
|
|
|| currentSecret.version !== freshness.version)) return;
|
|
const current = await aiConnectionService(
|
|
tx as unknown as Db,
|
|
).credential({ ...selection, grant });
|
|
if (antigravitySubscription) {
|
|
// Google refresh tokens have no provider freshness counter. Only
|
|
// the exact selected generation may replace itself; a reconnect,
|
|
// rotation, grant remap or concurrent refresh wins over this run.
|
|
if (sanitizeAntigravityCredential(current) !== originalCredential) return;
|
|
await secretService(tx).rotate(ref.secretId,
|
|
{ value: refreshed, preserveAiSessionEpoch: true }, { userId: grant.subjectUserId });
|
|
return;
|
|
}
|
|
const destination = path.join(
|
|
providerHome,
|
|
"current-auth.json",
|
|
);
|
|
await writeFile(destination, current, { mode: 0o600 });
|
|
const decision =
|
|
input.binding.provider === "openai"
|
|
? await decideCodexAuthMerge(authFile, destination, {
|
|
errorLabel: "AI account refresh",
|
|
})
|
|
: await decideGrokAuthMerge(authFile, destination, {
|
|
errorLabel: "AI account refresh",
|
|
});
|
|
if (decision !== 10) return;
|
|
await secretService(tx).rotate(
|
|
ref.secretId,
|
|
{ value: refreshed, preserveAiSessionEpoch: true },
|
|
{ userId: grant.subjectUserId },
|
|
);
|
|
// Keep grant.updatedAt for explicit account/access changes.
|
|
// The rotated secret revision invalidates quota caches without
|
|
// rejecting an otherwise valid in-flight reconnect.
|
|
}));
|
|
for (let attempt = 0; ; attempt++) {
|
|
try { await writeBack(); break; }
|
|
catch (error) {
|
|
if (attempt >= 2 || (error as NodeJS.ErrnoException)?.code !== WORKSPACE_RESTORE_LOCK_TIMEOUT_CODE) throw error;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
await googleRecovery?.clear();
|
|
// A committed save, intentional generation discard, or encrypted
|
|
// recovery handoff must own the refresh before deleting its home.
|
|
removalOnly = true;
|
|
await removeHome();
|
|
} catch (error) {
|
|
if (!durableGoogleRefresh && !removalOnly) throw error;
|
|
// Once native retirement and encrypted fsync transfer ownership, the
|
|
// original controller/home is unnecessary for next-use recovery.
|
|
if (!removalOnly) {
|
|
removalOnly = true;
|
|
deferredWriteback = { cause: error };
|
|
}
|
|
await removeHome();
|
|
if (deferredWriteback) throw new RetiredCredentialRecoveryError(deferredWriteback.cause);
|
|
}
|
|
};
|
|
// Journal writes and generation checks share the same reconnect fence.
|
|
try {
|
|
return await (antigravitySubscription
|
|
? withAccountHomeSecretMutationLock(undefined, input.companyId, cleanup) : cleanup());
|
|
} catch (error) {
|
|
if (retainedCleanupScope && !homeRemoved) retainAiCredentialCleanup({ scope: retainedCleanupScope, home: providerHome, cleanup: cleanupRuntime });
|
|
throw error;
|
|
}
|
|
})().finally(() => { cleanupRunning = undefined; });
|
|
return {
|
|
sessionIdentity,
|
|
...(retainedCleanupScope ? { retainedCleanupScope } : {}),
|
|
...(subscriptionFile && input.binding.provider === "openai" && credentialRef && freshness ? {
|
|
credentialWriteback: { grantId: selection.grant.id, secretId: credentialRef.secretId, sessionEpoch: freshness.epoch,
|
|
versionSelector: credentialRef.versionSelector ?? "latest" } satisfies PluginTaskCredentialWritebackBinding,
|
|
} : {}),
|
|
config: {
|
|
...input.config,
|
|
...projected?.config,
|
|
...(routing ? { managedAiRouting: routing } : {}),
|
|
env,
|
|
managedAiConnection: { ...attribution, identity, sessionIdentity },
|
|
},
|
|
attribution,
|
|
accountName: selection.connection.name,
|
|
accountOwnerUserId: selection.grant.subjectUserId,
|
|
identity,
|
|
home,
|
|
cleanup: cleanupRuntime,
|
|
};
|
|
} catch (error) {
|
|
if (home) await rm(home, { recursive: true, force: true });
|
|
releaseCredentialHome?.();
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
export async function withManagedAiProbe<T>(
|
|
db: Db,
|
|
input: Parameters<typeof prepareManagedAiRuntime>[1],
|
|
probe: (runtime: Awaited<ReturnType<typeof prepareManagedAiRuntime>>) => Promise<T>,
|
|
): Promise<T> {
|
|
const runtime = await prepareManagedAiRuntime(db, input);
|
|
let retained = false;
|
|
try {
|
|
return await probe(runtime);
|
|
}
|
|
catch (error) {
|
|
if (error instanceof NativeEnvironmentProbeCleanupError) {
|
|
// Keep the selected credential generation until the native owner has
|
|
// retired and completed its refresh handoff. HTTP need not wait for it.
|
|
retained = true;
|
|
const retryCleanup = (error as { retryCleanup?: () => Promise<void> }).retryCleanup;
|
|
retainAiCredentialCleanup({ scope: runtime.retainedCleanupScope!, home: runtime.home && path.join(runtime.home, "provider"), cleanup: runtime.cleanup, retired: error.retired,
|
|
...(typeof retryCleanup === "function" ? { retryRetirement: () => retryCleanup.call(error) } : {}) });
|
|
throw unprocessable("The native account test is still cleaning up. Retry after runtime cleanup completes.", { code: "native_probe_cleanup_pending" });
|
|
}
|
|
throw error;
|
|
} finally {
|
|
if (!retained) {
|
|
try { await runtime.cleanup(); }
|
|
catch (error) {
|
|
if (runtime.retainedCleanupScope && !(error instanceof RetiredCredentialRecoveryError)) {
|
|
retainAiCredentialCleanup({ scope: runtime.retainedCleanupScope, home: runtime.home && path.join(runtime.home, "provider"), cleanup: runtime.cleanup });
|
|
}
|
|
throw error;
|
|
}
|
|
}
|
|
}
|
|
}
|