eve

/** * Public, versioned summary of an eve agent emitted by `eve` * during `eve build`. * * This schema is the stable contract between eve and Vercel platform * surfaces (eg. the dashboard's project overview card). It is intentionally * smaller and more conservative than the internal {@link CompiledAgentManifest} * — only fields safe to render in product UI live here, and the schema is * versioned independently of the internal compiler manifest. * * **Storage** — the file is written to {@link VERCEL_EVE_AGENT_SUMMARY_OUTPUT_PATH} * (under the agent's `appRoot`, outside the Build Output API surface). On * Vercel deployments, the build container uploads it to a top-level S3 * artifact at `<projectId>/<deploymentId>/eve_agent_summary.json` — the * same tier as `deploy_metadata.json` and `turbo_summary.json`, never * reachable from the public CDN. The dashboard reads it through the * dedicated `/v6/deployments/:id/files/eve-agent-summary` endpoint with * team-scoped authentication. * * **Privacy** — non-public by storage location. The summary contains * user-authored business logic (tool descriptions, connection URLs, * source paths, subagent prompts) and is never served on the deployment * URL because the public proxy only resolves URLs against `basepath/`. * * **Self-hosted setups** — the file exists at the same path on disk * (`<appRoot>/.eve/agent-summary.json`) and is consumed by whatever * HTTP layer the operator wires up. */ /** * Stable schema kind embedded in every emitted summary file. */ export declare const VERCEL_EVE_AGENT_SUMMARY_KIND: "vercel-eve-agent-summary"; /** * Current public schema version. Bump only when adding required fields or * making semantic changes consumers must opt into. Adding optional fields * does not require a version bump. */ export declare const VERCEL_EVE_AGENT_SUMMARY_VERSION = 3; /** * Output path (relative to the agent's `appRoot`) where eve writes the * summary file at build time. * * Lives under `.eve/` — the existing eve-internal cache namespace that * already holds `nitro/`, `nitro-output/`, `sandbox-cache/`, etc. (see * `packages/eve/src/internal/application/paths.ts`). Intentionally * outside `.vercel/output/` so the file is not part of the Build Output * API surface — the Vercel build container picks it up from this path * and uploads it to a top-level deployment artifact, matching how the * Turbo run summary works. */ export declare const VERCEL_EVE_AGENT_SUMMARY_OUTPUT_PATH = ".eve/agent-summary.json"; /** * Display category eve exposes to the dashboard for one channel chip. Built * from the channel's reported {@link CompiledChannelDefinition.adapterKind}. */ export type VercelEveChannelType = "slack" | "http" | "webhook" | "unknown"; /** * Top-level agent identity used to label the dashboard card. */ export interface VercelEveAgentEntry { readonly name: string; readonly description?: string; readonly modelId: string; } /** * Authored agent instructions resolved at build time from the agent's * `instructions.md` or `instructions.{ts,cts,mts,js,cjs,mjs}` source. * Agents without authored instructions fall back to the framework default * and the summary's `instructions` field is `null`. * * The dashboard renders the markdown body verbatim, so the field carries * the full resolved content rather than a preview. For module-backed * sources the markdown is the result the module produced at build time, * not the module's source code. */ export interface VercelEveInstructionsEntry { /** * Logical path of the discovered instructions source, relative to the * agent root (eg. `instructions.md`). */ readonly logicalPath: string; /** * How the instructions were authored. `markdown` is a literal markdown * file at the agent root; `module` is a TypeScript / JavaScript module * that produces the instructions at build time. */ readonly sourceKind: "markdown" | "module"; /** Resolved markdown body of the instructions. */ readonly markdown: string; } export interface VercelEveScheduleEntry { readonly name: string; readonly cron: string; readonly logicalPath: string; } export interface VercelEveToolEntry { readonly name: string; readonly description: string; readonly logicalPath: string; } /** * Wire protocol the authored connection speaks, mirrored from the * compiled connection's `protocol` discriminator. `"mcp"` is a * {@link defineMcpClientConnection} server; `"openapi"` is a * {@link defineOpenAPIConnection} REST API. New primitives extend this * union when they ship. */ export type VercelEveConnectionType = "mcp" | "openapi"; export interface VercelEveConnectionEntry { readonly name: string; readonly description: string; readonly url: string; readonly logicalPath: string; readonly type: VercelEveConnectionType; /** * When the connection's auth is built by `connect()` from * `@vercel/connect/eve`, the connector identifier the author passed * (UID like `"oauth/mcp-linear-app"` or opaque `"scl_..."` form). * Dashboards use this to deep-link into the connector's settings page * — typically resolving UID → `scl_...` against * `GET /v6/connect/connectors` and building * `/[teamSlug]/[project]/connect/<scl_...>`. * * Omitted for connections backed by a raw MCP server, a static * token, or a custom `getToken` callback (anything not produced by * the Vercel Connect helper). */ readonly vercelConnect?: { readonly connector: string; }; } export interface VercelEveChannelEntry { readonly name: string; readonly method: "GET" | "POST" | "PUT" | "PATCH" | "DELETE" | "WEBSOCKET"; readonly urlPath: string; readonly type: VercelEveChannelType; /** * The raw `ChannelAdapter.kind` reported by the route, when present. * Useful when `type` is `"unknown"` and a consumer wants to render a * disambiguating label. */ readonly adapterKind?: string; readonly logicalPath: string; } export interface VercelEveSandboxEntry { readonly logicalPath: string; } /** * One authored skill exposed to the dashboard. * * Skills are markdown procedures the model loads on demand through the * framework-owned `load_skill` tool. The summary lists the available * skills by `name` + `description` so the dashboard can render the * surface without pulling each skill's full markdown body. Consumers * that need the full content can drill in through a dedicated endpoint * later — this entry is intentionally an overview shape. * * `sourceKind` distinguishes flat markdown skills (`agent/skills/foo.md`), * module-backed skills (`agent/skills/foo.ts`), and packaged skills * (`agent/skills/foo/SKILL.md` with sibling `references/`, `assets/`, * `scripts/` directories). */ export interface VercelEveSkillEntry { readonly name: string; readonly description: string; readonly logicalPath: string; readonly sourceKind: "markdown" | "module" | "skill-package"; } export interface VercelEveSubagentEntry { readonly name: string; readonly description: string; readonly logicalPath: string; } export interface VercelEveDiagnosticsSummary { readonly errors: number; readonly warnings: number; } /** * Versioned public summary of one eve agent, emitted into the Vercel Build * Output during `eve build` and ingested by Vercel as deployment metadata. */ export interface VercelEveAgentSummary { readonly kind: typeof VERCEL_EVE_AGENT_SUMMARY_KIND; readonly schemaVersion: typeof VERCEL_EVE_AGENT_SUMMARY_VERSION; /** * Version of the `eve` package that produced this summary, * for diagnostics. Format follows semver. */ readonly generatorVersion: string; readonly agent: VercelEveAgentEntry; /** * Authored agent instructions, when declared. `null` when the agent * relies on the framework default. */ readonly instructions: VercelEveInstructionsEntry | null; readonly schedules: readonly VercelEveScheduleEntry[]; readonly tools: readonly VercelEveToolEntry[]; readonly skills: readonly VercelEveSkillEntry[]; readonly connections: readonly VercelEveConnectionEntry[]; readonly channels: readonly VercelEveChannelEntry[]; readonly sandbox: VercelEveSandboxEntry | null; readonly subagents: readonly VercelEveSubagentEntry[]; readonly diagnostics: VercelEveDiagnosticsSummary; } /** * Normalizes a channel adapter's reported `kind` string to the closed set * the dashboard renders. Keeps the surface stable even when authors set a * custom `kind` on their adapter (eg. `"weather-slack"`). */ export declare function normalizeChannelKindForDisplay(adapterKind: string | undefined): VercelEveChannelType;