@cat-factory/node-server 0.107.6 → 0.107.8

package/dist/container.js

@@ -1,16 +1,14 @@
-import { AiAgentExecutor, defaultAgentKindRegistry, defaultInitiativePresetRegistry, inlineWebSearchOptionsFromEnv, resolveAgentConfig, isProxyableProvider, vendorConcurrencyLimiterFromEnv, } from '@cat-factory/agents';
+import { AiAgentExecutor, defaultAgentKindRegistry, defaultInitiativePresetRegistry, inlineWebSearchOptionsFromEnv, vendorConcurrencyLimiterFromEnv, } from '@cat-factory/agents';
 // Opt-in AWS EKS backends (runner + environment), registered by reference below (the Worker
 // facade registers the same pair, keeping the runtimes symmetric with the native `kubernetes`
 // backend these extend). They are pass-throughs until a workspace actually connects an `eks`
 // backend, and carry NO runtime AWS SDK dependency (the token is minted with WebCrypto), so this
 // adds no cost to a deployment that never uses EKS.
 import { eksEnvironmentBackend, eksRunnerBackend } from '@cat-factory/eks';
-import { ConfluenceProvider, FigmaProvider, ZeplinProvider, GitHubDocsProvider, GitHubIssuesProvider, JiraProvider, LinearDocumentProvider, LinearTaskProvider, createBackendRegistries, HttpRunnerPoolProvider, NotionProvider, ApiKeyService, LocalModelEndpointService, PersonalSubscriptionService, ProviderSubscriptionService, RunnerPoolConnectionService, ProvisioningLogRecorder, LoggingRunnerTransport, EMAIL_CIPHER_INFO, createEmailSender, SLACK_CIPHER_INFO, SlackNotificationChannel, TicketTrackerService, IssueWritebackService, githubIssuesLogic, createGitHubIssueViaToken, OBSERVABILITY_CIPHER_INFO, RegistryReleaseHealthProvider, defaultObservabilityRegistry, RegistrySubscriptionQuotaProvider, defaultSubscriptionQuotaRegistry, WorkspaceIncidentEnrichmentProvider, INCIDENT_ENRICHMENT_CIPHER_INFO, AccountSettingsService, ACCOUNT_SETTINGS_CIPHER_INFO, TestSecretsService, TEST_SECRETS_CIPHER_INFO, } from '@cat-factory/integrations';
-import { CompositeNotificationChannel, DEFAULT_MODEL_PRESET_ID, SUBSCRIPTION_VENDORS, composeTraceSinks, defaultProviderRegistry, defaultVcsRegistry, isAmbientNativeVendor, } from '@cat-factory/kernel';
+import { ConfluenceProvider, FigmaProvider, ZeplinProvider, GitHubDocsProvider, GitHubIssuesProvider, JiraProvider, LinearDocumentProvider, LinearTaskProvider, createBackendRegistries, HttpRunnerPoolProvider, NotionProvider, ApiKeyService, LocalModelEndpointService, ProvisioningLogRecorder, EMAIL_CIPHER_INFO, createEmailSender, SLACK_CIPHER_INFO, SlackNotificationChannel, TicketTrackerService, IssueWritebackService, githubIssuesLogic, OBSERVABILITY_CIPHER_INFO, RegistryReleaseHealthProvider, defaultObservabilityRegistry, RegistrySubscriptionQuotaProvider, defaultSubscriptionQuotaRegistry, WorkspaceIncidentEnrichmentProvider, INCIDENT_ENRICHMENT_CIPHER_INFO, AccountSettingsService, ACCOUNT_SETTINGS_CIPHER_INFO, TestSecretsService, TEST_SECRETS_CIPHER_INFO, } from '@cat-factory/integrations';
+import { CompositeNotificationChannel, DEFAULT_MODEL_PRESET_ID, defaultProviderRegistry, defaultVcsRegistry, } from '@cat-factory/kernel';
 import { AgentContextObservabilityService, SearchQueryObservabilityService, createCore, defaultStepResolverRegistry, PACKAGE_REGISTRY_CIPHER_INFO, resolvePackageRegistriesForDispatch, LlmObservabilityService, makeHarnessCallRecorder, resolvePresetModelForKind, } from '@cat-factory/orchestration';
-import { createLangfuseSink } from '@cat-factory/observability-langfuse';
-import { createNodeOtelSink } from '@cat-factory/observability-otel/node';
-import { CompositeAgentExecutor, ContainerAgentExecutor, ContainerEnvConfigRepairer, ContainerRepoBootstrapper, ContainerSessionService, FanOutEventPublisher, FetchGitHubClient, FetchGitHubProvisioningClient, GitHubAppAuth, GitHubAppRegistry, GitHubIdentityResolver, GitHubCiStatusProvider, GitHubDocQualityProvider, GitHubMergeabilityProvider, GitHubPullRequestReviewProvider, GitHubBranchUpdater, GitHubPullRequestMerger, InAppNotificationChannel, PatPreferringAppRegistry, runWithInitiator, WebCryptoPasswordHasher, WebCryptoSecretCipher, WebCryptoWebhookVerifier, buildInfrastructureCapabilities, testEnvHasZeroConfigDefault, buildResolveRepoTarget, buildResolveRepoTargets, createDefaultWebSearchUpstream, createWebSearchUpstream, makePreviewJobBuilder, makeResolveDeployCloneTarget, makeResolveRunRepoContext, makeResolveRepoFilesForCoords, makeResolveBinaryArtifactStore, RunnerJobClient, DOCS, ENV_VARS_ANCHORS, noRunnerBackendAvailableError, ensureWorkBranchViaRest, logger, resolveUrlSafetyPolicy, resolveWorkspaceCapabilities, wrapResolverWithLimiter, } from '@cat-factory/server';
+import { CompositeAgentExecutor, ContainerSessionService, FanOutEventPublisher, FetchGitHubClient, FetchGitHubProvisioningClient, GitHubAppAuth, GitHubAppRegistry, GitHubIdentityResolver, GitHubCiStatusProvider, GitHubDocQualityProvider, GitHubMergeabilityProvider, GitHubPullRequestReviewProvider, GitHubBranchUpdater, GitHubPullRequestMerger, InAppNotificationChannel, PatPreferringAppRegistry, runWithInitiator, WebCryptoPasswordHasher, WebCryptoSecretCipher, WebCryptoWebhookVerifier, buildInfrastructureCapabilities, testEnvHasZeroConfigDefault, buildResolveRepoTarget, buildResolveRepoTargets, createDefaultWebSearchUpstream, createWebSearchUpstream, makePreviewJobBuilder, makeResolveDeployCloneTarget, makeResolveRunRepoContext, makeResolveRepoFilesForCoords, makeResolveBinaryArtifactStore, RunnerJobClient, logger, resolveUrlSafetyPolicy, resolveWorkspaceCapabilities, wrapResolverWithLimiter, } from '@cat-factory/server';
 // The built-in polling-gate suite (ci / conflicts / post-release-health + on-call). The facade
 // builds an app-owned `GateRegistry` pre-loaded with the suite via `gateRegistryWithBuiltins()`
 // below, then wires each gate's provider.
@@ -47,9 +45,12 @@ import { DrizzleSlackConnectionRepository, DrizzleSlackMemberMappingRepository,
 import { DrizzleTaskConnectionRepository, DrizzleTaskRepository, DrizzleTaskSourceSettingsRepository, } from './repositories/tasks.js';
 import { CryptoIdGenerator, SystemClock } from './runtime.js';
 import { buildNodeApiKeyService, buildNodeLocalModelEndpointService, buildNodeOpenRouterCatalogService, buildNodePersonalSubscriptionService, buildNodePublicApiKeyService, buildNodeSubscriptionService, buildNodeUserSecretService, } from './wireCredentialServices.js';
-// HKDF domain tag separating runner-pool scheduler secrets from any other use of
-// the same master key (mirrors the Worker's `cat-factory:runners`).
-const RUNNERS_CIPHER_INFO = 'cat-factory:runners';
+// The container-agent-executor wiring (transport resolver, provisioning-log wrapper, container
+// executor + bootstrapper + env-config repairer, GitHub-issue filer, trace-sink builder), lifted
+// into a sibling module so this composition root stays within the file-size budget.
+import { RUNNERS_CIPHER_INFO, buildNodeContainerExecutor, buildNodeGitHubIssueFiler, buildNodeResolveTransport, buildTraceSink, selectNodeEnvConfigRepairer, selectNodeRepoBootstrapper, withProvisioningLog, } from './container-executor-deps.js';
+// Re-export the public seams the local facade + tests still import from `./container.js`.
+export { buildNodeResolveTransport, missingContainerExecutorPrereqs, withProvisioningLog, } from './container-executor-deps.js';
 /**
  * Source one org/durable repository that a standard build constructs directly from the Drizzle
  * `db`. In mothership mode (no Postgres) `remote` is the full-surface remote registry — a
@@ -71,7 +72,6 @@ export function pickRepoSource(remote, name, build) {
 // own. Mirrors the Worker's `buildModelProvider` memoisation. Memoisation matters more for
 // OTel than Langfuse: the SDK sink owns batch processors/exporters, so it must be built
 // once per config, not per wiring site.
-const traceSinkCache = new WeakMap();
 /** Truthy env flag (`true`/`1`/`yes`). */
 function isTruthy(value) {
     return value === 'true' || value === '1' || value === 'yes';
@@ -98,38 +98,6 @@ instrument) {
     modelResolverCache.set(db, resolver);
     return resolver;
 }
-/**
- * Build the opt-in external trace sink(s) — Langfuse and/or OpenTelemetry — composed into
- * the single sink slot; the observability service then fans every recorded LLM call out to
- * whichever are wired. Memoised per config so both wiring sites share one instance.
- *
- * Langfuse uses the fetch-based sink (identical to the Worker). OpenTelemetry uses the
- * OFFICIAL `@opentelemetry/*` SDK exporter (`createNodeOtelSink`) — the Node counterpart of
- * the Worker's fetch OTLP exporter, kept conformant by the shared mapping layer + tests.
- */
-function buildTraceSink(config) {
-    if (traceSinkCache.has(config))
-        return traceSinkCache.get(config);
-    const langfuse = !config.langfuse.enabled || !config.langfuse.publicKey || !config.langfuse.secretKey
-        ? undefined
-        : createLangfuseSink({
-            publicKey: config.langfuse.publicKey,
-            secretKey: config.langfuse.secretKey,
-            baseUrl: config.langfuse.baseUrl,
-            logger,
-        });
-    const otel = !config.otel.enabled || !config.otel.endpoint
-        ? undefined
-        : createNodeOtelSink({
-            endpoint: config.otel.endpoint,
-            headers: config.otel.headers,
-            serviceName: config.otel.serviceName,
-            logger,
-        });
-    const sink = composeTraceSinks([langfuse, otel]);
-    traceSinkCache.set(config, sink);
-    return sink;
-}
 /**
  * Wire the Slack integration when enabled: the notification *channel* (an extra
  * delivery transport composed onto the notification mechanism — Node has no in-app
@@ -297,358 +265,6 @@ function buildNodeVcsIdentityRegistry(config) {
  * when runner pools are not enabled. Mirrors the Worker's `buildResolveTransport`,
  * minus the Cloudflare-container path.
  */
-export function buildNodeResolveTransport(config, 
-// The port, not the Drizzle concrete: in mothership mode the local facade passes a remote
-// (RPC-backed) connection repo, and the service layer only ever uses the port methods.
-runnerPoolConnectionRepository, workspaceRepository, clock, 
-// The app-owned runner-backend registry the service resolves a stored `kind` through.
-runnerBackendRegistry, 
-// The shared HTTP provider the built-in `manifest` backend reuses when supplied (e.g.
-// tests). NOT the custom-kind seam — a bespoke runner backend is registered by reference
-// into `runnerBackendRegistry`. Absent → the generic manifest-driven HTTP provider.
-injectedPoolProvider) {
-    if (!config.runners.enabled || !config.runners.encryptionKey)
-        return null;
-    const urlPolicy = resolveUrlSafetyPolicy(config.runners);
-    const runnerService = new RunnerPoolConnectionService({
-        runnerPoolConnectionRepository,
-        workspaceRepository,
-        secretCipher: new WebCryptoSecretCipher({
-            masterKeyBase64: config.runners.encryptionKey,
-            info: RUNNERS_CIPHER_INFO,
-        }),
-        clock,
-        runnerBackendRegistry,
-        ...(urlPolicy ? { urlPolicy } : {}),
-        runnerPoolProvider: injectedPoolProvider ?? new HttpRunnerPoolProvider(urlPolicy ? { urlPolicy } : {}),
-    });
-    return async (workspaceId) => {
-        if (workspaceId) {
-            const resolved = await runnerService.resolve(workspaceId);
-            if (resolved)
-                return resolved.transport;
-        }
-        // The shared factory throws a ConflictError carrying the machine reason (see its doc): a clean
-        // 409 synchronously, and classifyDispatchFailure lifts the reason onto the run's AgentFailure on
-        // the async dispatch path (SPA shows "Agent backend not configured", not "container failed to
-        // start"). The Node facade has no per-run container backend, so the remedy points only at the
-        // self-hosted runner pool / Kubernetes.
-        throw noRunnerBackendAvailableError(workspaceId);
-    };
-}
-/**
- * Wrap a transport resolver so every dispatch/release/poll-failure appends a
- * provisioning-log event. A no-op when there's no resolver. `subsystem` tags the
- * rows (a self-hosted pool vs a per-run container) so the logs drawer can filter.
- */
-export function withProvisioningLog(resolve, recorder, subsystem) {
-    if (!resolve)
-        return null;
-    // Closure-owned so it survives each (per-resolution) wrapper: a terminal `failed`
-    // job re-polled by a replay/re-drive logs its poll-failure only once.
-    const loggedPollFailures = new Set();
-    return async (workspaceId) => {
-        const inner = await resolve(workspaceId);
-        return new LoggingRunnerTransport({
-            inner,
-            recorder,
-            workspaceId: workspaceId ?? '',
-            subsystem,
-            loggedPollFailures,
-        });
-    };
-}
-/**
- * Which of the container-executor prerequisites are missing, as the human labels the boot
- * warning names. Empty ⇒ all three are present. `PUBLIC_URL` is this service's externally
- * reachable base backing the LLM proxy, `AUTH_SESSION_SECRET` signs the harness↔proxy tokens,
- * and a runner backend is what a dispatch is handed to. Pure so the "name exactly what's
- * missing" logic is unit-tested (error-message coverage A5).
- */
-export function missingContainerExecutorPrereqs(input) {
-    const missing = [];
-    if (!input.publicUrl)
-        missing.push('PUBLIC_URL');
-    if (!input.sessionSecret)
-        missing.push('AUTH_SESSION_SECRET (>= 32 chars)');
-    if (!input.hasRunnerBackend)
-        missing.push('a runner backend (self-hosted runner pool)');
-    return missing;
-}
-/**
- * Build the container agent executor (repo-operating steps: coder, mocker,
- * playwright, blueprints, ci-fixer, conflict-resolver, merger) when its
- * prerequisites are configured: a token source for the push/clone token, the public
- * URL backing the LLM proxy, the session secret to sign proxy tokens, and a runner
- * backend. Returns null when any is missing, so the composite fails those kinds
- * loudly rather than running them as useless one-shot LLM calls.
- *
- * The token source is pluggable: a sibling facade may pass `mintInstallationToken`
- * (e.g. a static PAT for local mode), otherwise it is minted via the GitHub App
- * registry (which additionally requires the App private key + `github.enabled`).
- */
-function buildNodeContainerExecutor(env, config, appRegistry, resolveRepoTarget, resolveRepoTargets, resolveTransport, resolveWorkspaceModelDefault, agentKindRegistry, mintInstallationTokenOverride, subscriptions, personalSubscriptions, resolveAccountId, resolveUserGitHubToken, agentContextObservability, resolveWebSearchAvailability, resolveRepoOrigin, resolvePackageRegistries, resolveTestSecrets, recordHarnessCalls, recordSubscriptionQuotaUsage) {
-    // The harness reaches models only through this service's LLM proxy; `PUBLIC_URL`
-    // is this service's externally reachable base (the runner pool / local container
-    // must be able to reach it). Pi posts to `${PUBLIC_URL}/v1/chat/completions`.
-    const publicUrl = env.PUBLIC_URL?.trim();
-    const sessionSecret = config.auth.sessionSecret;
-    if (!publicUrl || !sessionSecret || !resolveTransport) {
-        // The executor is disabled but the service still boots "healthy" — repo-operating steps
-        // (coder/mocker/tester/blueprints/ci-fixer/conflict-resolver/merger) then fail only at
-        // dispatch, deep in a request, with no boot signal. Emit a greppable line naming exactly
-        // which prerequisite is missing so the gap is visible up front (error-message coverage A5).
-        const missing = missingContainerExecutorPrereqs({
-            publicUrl,
-            sessionSecret,
-            hasRunnerBackend: !!resolveTransport,
-        });
-        logger.warn({ missing, docsUrl: DOCS.envVars(ENV_VARS_ANCHORS.coreServiceNetworking) }, `container agent steps are DISABLED: missing ${missing.join(', ')}. Repo-operating steps ` +
-            `(coder/mocker/tester/merger/…) will fail at dispatch until configured. See ` +
-            `${DOCS.envVars(ENV_VARS_ANCHORS.coreServiceNetworking)}.`);
-        return null;
-    }
-    // Token source: an explicit override (e.g. a static PAT in local mode) wins; else
-    // the GitHub App registry mints a per-installation token (when the App is configured).
-    const baseMint = mintInstallationTokenOverride ??
-        (appRegistry ? (id) => appRegistry.installationToken(id) : undefined);
-    if (!baseMint) {
-        // Every other prerequisite is set but there is no GitHub token source, so the harness
-        // could never clone/push. Name the fix (App creds) rather than disabling silently (A5).
-        logger.warn({ missing: ['GITHUB_APP_ID + GITHUB_APP_PRIVATE_KEY'], docsUrl: DOCS.githubOperations() }, `container agent steps are DISABLED: no GitHub token source — set GITHUB_APP_ID + ` +
-            `GITHUB_APP_PRIVATE_KEY so the harness can mint a push/clone token. Repo-operating steps ` +
-            `will fail at dispatch until configured. See ${DOCS.githubOperations()}.`);
-        return null;
-    }
-    // Prefer the run initiator's per-user PAT (when stored) over the App/env token, so
-    // pushes/PRs are attributed to them. Falls back to the base mint otherwise.
-    const mintInstallationToken = async (installationId, ctx) => {
-        if (resolveUserGitHubToken && ctx?.initiatedBy) {
-            const pat = await resolveUserGitHubToken(ctx.initiatedBy);
-            if (pat)
-                return pat;
-        }
-        return baseMint(installationId);
-    };
-    return new ContainerAgentExecutor({
-        resolveTransport,
-        agentRouting: config.agents.routing,
-        resolveBlockModel: config.agents.resolveBlockModel,
-        resolveWorkspaceModelDefault,
-        resolveRepoTarget,
-        // Multi-repo coding (service-connections phase 3): the implementer fans a cross-service
-        // change out across the task's own repo + each connected involved-service repo.
-        resolveRepoTargets,
-        ...(resolveAccountId ? { resolveAccountId } : {}),
-        mintInstallationToken,
-        // Ensure the shared per-task work branch up front so every agent (including the
-        // read-only architect) operates on the same branch — idempotent, best-effort. Writers
-        // create it from base; read-only agents only probe (`options.create`).
-        ensureWorkBranch: async (repo, branch, options) => ensureWorkBranchViaRest({
-            ...(config.github.apiBase ? { apiBase: config.github.apiBase } : {}),
-            token: await mintInstallationToken(repo.installationId),
-            owner: repo.owner,
-            name: repo.name,
-            baseBranch: repo.baseBranch,
-            branch,
-            create: options.create,
-        }),
-        sessionService: new ContainerSessionService({ secret: sessionSecret }),
-        // The subscription harnesses (Claude Code / Codex) lease a pooled token and
-        // attribute usage back for usage-aware rotation; absent ⇒ those harnesses are
-        // unavailable and a subscription-only model fails loudly at dispatch.
-        ...(subscriptions
-            ? {
-                leaseSubscriptionToken: (workspaceId, vendor) => subscriptions.leaseToken(workspaceId, vendor),
-                recordSubscriptionUsage: (workspaceId, tokenId, usage) => subscriptions.recordTokenUsage(workspaceId, tokenId, usage),
-                hasSubscriptionToken: (workspaceId, vendor) => subscriptions.hasToken(workspaceId, vendor),
-            }
-            : {}),
-        // Per-call telemetry for the subscription harnesses (proxy-bypassing), recorded
-        // into `llm_call_metrics` alongside the proxy-metered Pi rows.
-        ...(recordHarnessCalls ? { recordHarnessCalls } : {}),
-        // Modeled subscription quota-cycle tracking (Part B): fold a finished subscription
-        // run's tokens into the rolling windows, for BOTH pooled and personal runs.
-        ...(recordSubscriptionQuotaUsage ? { recordSubscriptionQuotaUsage } : {}),
-        // Individual-usage harnesses (Claude) lease the run-initiator's OWN activated
-        // personal credential; absent ⇒ such models fail loudly at dispatch.
-        ...(personalSubscriptions
-            ? {
-                leasePersonalSubscriptionToken: (executionId, userId, vendor) => personalSubscriptions.leaseForRun(executionId, userId, vendor),
-                // Route a dual-mode individual model (GLM) to the initiator's own subscription
-                // when they have one; otherwise dispatch keeps it on the Cloudflare base.
-                hasPersonalSubscription: (userId, vendor) => personalSubscriptions.has(userId, vendor),
-            }
-            : {}),
-        // Native local execution (local facade, opt-in): run subscription-harness agents with
-        // the developer's OWN installed CLI + ambient login instead of leasing a credential.
-        // Ambient auth applies ONLY when the resolved harness is in the allow-list AND the
-        // vendor is that CLI's NATIVE vendor (no Anthropic-compatible base URL of its own:
-        // `claude` / `codex`). A non-native vendor reusing the `claude-code` harness
-        // (GLM/Kimi/DeepSeek carries its own `baseUrl`) is leased normally — otherwise ambient
-        // auth would silently drop that base URL and run the step on the developer's own
-        // Anthropic login instead of the pinned vendor.
-        ...(config.nativeAmbientAuth && config.nativeAmbientAuth.length > 0
-            ? {
-                // The allow-list + no-`baseUrl` check is the shared `isAmbientNativeVendor`
-                // predicate (so this can't drift from the personal-credential gate); the extra
-                // `harness === h` guard ensures the RESOLVED harness matches the vendor's own.
-                nativeAmbientAuth: (h, vendor) => vendor !== undefined &&
-                    SUBSCRIPTION_VENDORS[vendor].harness === h &&
-                    isAmbientNativeVendor(config.nativeAmbientAuth, vendor),
-            }
-            : {}),
-        proxyBaseUrl: `${publicUrl.replace(/\/+$/, '')}/v1`,
-        // Point container agents' web search at the backend search proxy (no provider key in
-        // the sandbox), but only for a run whose account has keys (resolved per run — see the
-        // call site), so the tool is never advertised to a run where it would just fail.
-        ...(resolveWebSearchAvailability ? { resolveWebSearchAvailability } : {}),
-        // Decrypt the workspace's private-registry entries onto the job body (rendered by
-        // the harness into ~/.npmrc), so private dependencies resolve on install.
-        ...(resolvePackageRegistries ? { resolvePackageRegistries } : {}),
-        // Decrypt the service frame's SENSITIVE test credentials onto the tester job body (out of
-        // band — injected as container env vars by the harness, never in the prompt/telemetry).
-        ...(resolveTestSecrets ? { resolveTestSecrets } : {}),
-        githubApiBase: config.github.apiBase,
-        // Resolve the clone URL + provider per repo. The local GitLab facade injects a GitLab
-        // origin so containers clone gitlab.com (or a self-managed host) and open MRs; absent ⇒
-        // the default github.com origin.
-        ...(resolveRepoOrigin ? { resolveRepoOrigin } : {}),
-        // Forward container tool spans to the external trace sink(s) (Langfuse and/or OTLP)
-        // grouped under the run trace — the same sink the LLM proxy fans generations to.
-        // (Langfuse nests them as children; the OTLP exporter groups them by shared trace id.)
-        llmTraceSink: buildTraceSink(config),
-        // Record the complete provided context per dispatch (best-effort, gated in the sink).
-        ...(agentContextObservability ? { agentContextObservability } : {}),
-        agentKindRegistry,
-    });
-}
-/**
- * Build the repo bootstrapper (the "bootstrap repo" container dispatch) when its
- * prerequisites are configured — mirroring the Worker's `selectRepoBootstrapper` and
- * the container-executor prerequisites: a resolvable runner transport, the public URL
- * + session secret backing the LLM proxy, a token source, and a GitHub client.
- * Returns undefined otherwise (the bootstrap module then has no runner and the service
- * reports a clean dispatch failure). Bootstrap is an `architect`-kind run, so it
- * follows that kind's routing. The promoted `ContainerRepoBootstrapper` dispatches
- * through the same shared runner seam the container executor uses, so on Node it runs
- * against the self-hosted pool and on local against the per-job Docker container.
- */
-function selectNodeRepoBootstrapper(deps) {
-    const publicUrl = deps.env.PUBLIC_URL?.trim();
-    const sessionSecret = deps.config.auth.sessionSecret;
-    if (!deps.resolveTransport ||
-        !publicUrl ||
-        !sessionSecret ||
-        !deps.githubClient ||
-        !deps.mintInstallationToken) {
-        return undefined;
-    }
-    return new ContainerRepoBootstrapper({
-        resolveTransport: deps.resolveTransport,
-        installationRepository: deps.installationRepository,
-        bootstrapJobRepository: deps.bootstrapJobRepository,
-        repoRepository: deps.repoRepository,
-        ...(deps.repoProjectionCache ? { repoProjectionCache: deps.repoProjectionCache } : {}),
-        githubClient: deps.githubClient,
-        mintInstallationToken: deps.mintInstallationToken,
-        sessionService: new ContainerSessionService({ secret: sessionSecret }),
-        model: resolveAgentConfig(deps.config.agents.routing, 'architect').ref,
-        proxyBaseUrl: `${publicUrl.replace(/\/+$/, '')}/v1`,
-        githubApiBase: deps.config.github.apiBase,
-        // The scaffolder installs dependencies too — forward the workspace's
-        // private-registry entries exactly as the implementation executor does.
-        ...(deps.resolvePackageRegistries
-            ? { resolvePackageRegistries: deps.resolvePackageRegistries }
-            : {}),
-    });
-}
-/**
- * Build the live ENVIRONMENT-PROVIDER CONFIG REPAIR agent (PR #416 increment 2) when its
- * prerequisites are met — the same container prerequisites as the bootstrapper PLUS a
- * registered backend that supports agent repair (`describeRepairAgent`). The stock manifest
- * provider has no repair support, so this stays undefined there; it wires only when a custom
- * backend registered into the env-backend registry implements repair (so local inherits it
- * too). NOT the repo bootstrapper: an ordinary clone→edit→push coding job, no history reset.
- */
-function selectNodeEnvConfigRepairer(deps) {
-    const publicUrl = deps.env.PUBLIC_URL?.trim();
-    const sessionSecret = deps.config.auth.sessionSecret;
-    // Prefer the internal override (the conformance suite's fake repair provider), else scan
-    // the env-backend registry for the first repair-capable backend. Built-ins don't support
-    // repair, so this is undefined on a stock deployment; a third-party backend wires it.
-    const repairUrlPolicy = resolveUrlSafetyPolicy(deps.config.environments);
-    const environmentProvider = !deps.resolveTransport
-        ? undefined
-        : (deps.override ??
-            deps.environmentBackendRegistry.findRepairCapable(repairUrlPolicy ? { urlPolicy: repairUrlPolicy } : {}));
-    if (!deps.resolveTransport ||
-        !publicUrl ||
-        !sessionSecret ||
-        !deps.mintInstallationToken ||
-        !environmentProvider ||
-        typeof environmentProvider.describeRepairAgent !== 'function') {
-        return undefined;
-    }
-    // A config fix is coding work, so it follows the `coder` kind's routing. The repair runs on
-    // the Pi harness over the LLM proxy, so the routed model MUST be proxyable. Surface a
-    // misconfiguration HERE (at wiring) rather than letting every repair dispatch throw deep in a
-    // request: if `coder` is routed to a non-proxyable model (e.g. an individual subscription
-    // vendor), leave the fallback unwired — bootstrap then returns the validation issues, exactly
-    // as it does when no provider supports repair.
-    const model = resolveAgentConfig(deps.config.agents.routing, 'coder').ref;
-    if (!isProxyableProvider(model.provider)) {
-        logger.warn({ provider: model.provider }, 'env-config repair: the coder routing model is not proxyable by the LLM proxy; ' +
-            'the agent config-repair fallback is disabled.');
-        return undefined;
-    }
-    return new ContainerEnvConfigRepairer({
-        resolveTransport: deps.resolveTransport,
-        installationRepository: deps.installationRepository,
-        mintInstallationToken: deps.mintInstallationToken,
-        sessionService: new ContainerSessionService({ secret: sessionSecret }),
-        environmentProvider,
-        model,
-        proxyBaseUrl: `${publicUrl.replace(/\/+$/, '')}/v1`,
-        githubApiBase: deps.config.github.apiBase,
-    });
-}
-/**
- * Build the GitHub-issue tracker filer for the tech-debt pipeline when the GitHub
- * App is configured. It resolves the service's repo from the workspace's
- * `github_repos` projection and mints a short-lived token from that workspace's OWN
- * App installation (per-tenant) — the same infra the container executor uses — then
- * files the issue via the token. Returns undefined when the App isn't configured (the
- * GitHub tracker then passes through). A run whose service isn't linked to a repo
- * resolves to null (a clean pass-through, not a run failure).
- */
-function buildNodeGitHubIssueFiler(config, registry, resolveRepoTarget) {
-    if (!registry)
-        return undefined;
-    return async (request) => {
-        let repo;
-        try {
-            repo = await resolveRepoTarget(request.workspaceId, request.frameId);
-        }
-        catch {
-            // The service isn't linked to a repo — nothing to file against; pass through.
-            return null;
-        }
-        if (!repo)
-            return null;
-        const token = await registry.installationToken(repo.installationId);
-        const issue = await createGitHubIssueViaToken({
-            fetchImpl: fetch,
-            token,
-            owner: repo.owner,
-            repo: repo.name,
-            title: request.title,
-            body: request.body,
-            apiBase: config.github.apiBase,
-        });
-        return { externalId: `${repo.owner}/${repo.name}#${issue.number}`, url: issue.url };
-    };
-}
 /**
  * The Node composition root: assemble the framework-agnostic domain `Core` with
  * Drizzle/Postgres repositories + Node implementations of the runtime ports, then