@boardwalk-labs/engine 0.1.0

package/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -0,0 +1,69 @@
+# boardwalk
+
+The open-source single-node engine for Boardwalk workflows: cron scheduling, durable run
+semantics, SQLite run history, and a local run log — on hardware you own, with no account.
+
+> **Status: pre-release.** This repo is being built in the open ahead of its first published
+> release. The contracts it implements are stable (see [`@boardwalk-labs/workflow`](https://www.npmjs.com/package/@boardwalk-labs/workflow));
+> the engine itself is under active construction. See [`SPEC.md`](./SPEC.md) for the
+> architecture and the definition of done.
+
+## What it is
+
+A workflow is a plain TypeScript program. This engine runs it:
+
+- **Server mode** — `docker run ghcr.io/boardwalk-labs/boardwalk`: a long-lived process that schedules cron
+  workflows, accepts webhooks, keeps run history in SQLite, and serves a local run log.
+- **Embedded mode** — `@boardwalk-labs/engine` as a library: one run, in-process supervision; this is
+  what `boardwalk dev` uses.
+
+Same engine, same semantics as the Boardwalk platform: one run = one process, `sleep` holds the
+process, a crash restarts the run from the top, `workflows.call` re-attaches idempotently.
+The conformance suite in this repo tests that parity.
+
+## Quickstart
+
+Run the server with Docker — run history and state live in the mounted data dir:
+
+```sh
+docker run -v ./data:/data -p 8080:8080 ghcr.io/boardwalk-labs/boardwalk
+```
+
+Then open `http://localhost:8080` for the run log, or hit the JSON API
+(`/api/workflows`, `/api/runs`). Webhook triggers land on `/hooks/<workflow>/<trigger-id>`.
+
+### Configuration
+
+All configuration is environment variables (a `boardwalk.toml` file is deferred — see
+[`SPEC.md`](./SPEC.md) §2.4):
+
+| Variable                  | Default                                    | What it does                                                                                 |
+| ------------------------- | ------------------------------------------ | -------------------------------------------------------------------------------------------- |
+| `BOARDWALK_DATA_DIR`      | `/data` in Docker, else `./boardwalk-data` | Where everything lives: SQLite DB, run dirs, artifacts                                       |
+| `BOARDWALK_HOST`          | `127.0.0.1` (`0.0.0.0` in Docker)          | Bind address — this surface has no auth beyond webhook auth, so binding wider logs a warning |
+| `BOARDWALK_PORT`          | `8080`                                     | Listen port (`0` picks a free port)                                                          |
+| `BOARDWALK_DEFAULT_MODEL` | —                                          | Model used when `agent()` omits one, e.g. `anthropic/claude-sonnet-4-5`                      |
+| `BOARDWALK_PROVIDERS`     | —                                          | JSON provider table, e.g. `{"ollama":{"base_url":"http://localhost:11434/v1"}}`              |
+| `BOARDWALK_ENV_FILE`      | `<data-dir>/.env`, if it exists            | `.env` file backing `secrets.get` and provider API keys                                      |
+
+Provider API keys come from the environment (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`,
+`GEMINI_API_KEY` for the built-ins; `api_key_env` names the variable for custom providers).
+
+### Embedded mode
+
+The same engine as a library — this is what `boardwalk dev` does:
+
+```ts
+import { Engine } from "@boardwalk-labs/engine";
+
+const engine = new Engine({ dataDir: "./boardwalk-data" });
+const run = await engine.runOnce({ program: bundledProgramSource });
+console.log(run.status, run.output);
+engine.close();
+```
+
+For OAuth-protected MCP servers an `agent()` call connects to, `engine.authorizeMcpServer(url, { onAuthorizationUrl })` performs the one-time interactive grant; after that, runs use (and silently refresh) the stored token headlessly — see [SPEC.md §2.3](./SPEC.md).
+
+## License
+
+Apache-2.0 — see [LICENSE](./LICENSE).

package/bin/boardwalk-server.js

@@ -0,0 +1,16 @@
+#!/usr/bin/env node
+// Thin launcher for `boardwalk-server` (SPEC §5). All real logic lives in compiled,
+// type-checked, tested TypeScript (src/server_main.ts) — this shim only exists so npm `bin`
+// and the Docker CMD share one entrypoint that resolves dist/ relative to the package.
+import("../dist/server_main.js")
+  .then((mod) => mod.main())
+  .catch((err) => {
+    // Boot failures are usually operator config mistakes: print the message (and the
+    // EngineError hint when present), never a stack trace into engine internals.
+    const message = err instanceof Error ? err.message : String(err);
+    process.stderr.write(`boardwalk-server: ${message}\n`);
+    if (err !== null && typeof err === "object" && typeof err.hint === "string") {
+      process.stderr.write(`  hint: ${err.hint}\n`);
+    }
+    process.exit(1);
+  });

package/dist/agent/conversation.d.ts

@@ -0,0 +1,42 @@
+/** A tool the model may call, as advertised to the provider. */
+export interface ToolSpec {
+    name: string;
+    description: string;
+    inputSchema: Record<string, unknown>;
+}
+/** A tool invocation the model requested in an assistant turn. */
+export interface ToolCallRequest {
+    /** Provider-assigned call id — echoed back with the result. */
+    id: string;
+    name: string;
+    input: Record<string, unknown>;
+}
+export type ChatMessage = {
+    role: "user";
+    text: string;
+} | {
+    role: "assistant";
+    text: string;
+    toolCalls: readonly ToolCallRequest[];
+} | {
+    role: "tool_results";
+    results: readonly ToolResultMessage[];
+};
+export interface ToolResultMessage {
+    /** The ToolCallRequest id this answers. */
+    id: string;
+    /** Result content, already stringified (and redacted) by the loop. */
+    content: string;
+    isError: boolean;
+}
+/** One model turn: final text and/or tool-call requests, plus usage. */
+export interface ChatTurn {
+    text: string;
+    toolCalls: ToolCallRequest[];
+    usage: {
+        inputTokens?: number | undefined;
+        outputTokens?: number | undefined;
+    };
+    /** True when the model stopped to call tools (the loop continues). */
+    wantsTools: boolean;
+}

package/dist/agent/conversation.js

@@ -0,0 +1,4 @@
+// The provider-neutral conversation model for the agent() loop. The leaf builds these;
+// each protocol adapter maps them to its wire format. Keeping the loop neutral is what lets
+// two adapters (Anthropic + OpenAI-compatible) cover every supported endpoint.
+export {};

package/dist/agent/leaf.d.ts

@@ -0,0 +1,81 @@
+import type { AgentOptions, TokenUsage, ToolReturn } from "@boardwalk-labs/workflow";
+import { type ProviderIo } from "./providers.js";
+import type { Redactor } from "./redact.js";
+import type { ResolvedModel } from "./resolve.js";
+import { type McpTokenResult, type ToolSetContext } from "./tools.js";
+/**
+ * Identity of one `agent()` leaf, carried on its `turn_started`/`turn_ended` frames so a stream
+ * consumer can tell concurrent agents apart. `agentId` is stable and run-unique (engine-assigned);
+ * `agentName` is the author's `AgentOptions.name`, present only when they set one.
+ */
+export interface AgentIdentity {
+    agentId: string;
+    agentName?: string;
+}
+/**
+ * A leaf event body, scoped to the leaf's turn. `turn_started` itself is emitted by the
+ * supervisor when the turn opens (io.startTurn) — the leaf only emits what follows.
+ */
+export type LeafEventBody = ({
+    kind: "turn_ended";
+    reason: "complete" | "error";
+    usage?: TokenUsage;
+    error?: {
+        code: string;
+        message: string;
+    };
+} & AgentIdentity) | {
+    kind: "text_start";
+    blockId: string;
+} | {
+    kind: "text_delta";
+    blockId: string;
+    text: string;
+} | {
+    kind: "text_end";
+    blockId: string;
+} | {
+    kind: "tool_call_start";
+    toolCallId: string;
+    toolName: string;
+} | {
+    kind: "tool_call_input_complete";
+    toolCallId: string;
+    input: Record<string, unknown>;
+} | {
+    kind: "tool_call_executing";
+    toolCallId: string;
+} | {
+    kind: "tool_call_result";
+    toolCallId: string;
+    result: ToolReturn;
+} | {
+    kind: "tool_call_error";
+    toolCallId: string;
+    error: {
+        code: string;
+        message: string;
+    };
+};
+export interface LeafIo {
+    /** This leaf's identity — stamped onto its turn_started (via startTurn) and turn_ended frames. */
+    identity: AgentIdentity;
+    /** Supervisor-side model resolution (config + key material never live in this process). */
+    resolve(model: string | undefined, provider: string | undefined): Promise<ResolvedModel>;
+    /** Open a new turn block; subsequent leaf events carry this turnId. */
+    startTurn(turnId: string): void;
+    emit(turnId: string, body: LeafEventBody): void;
+    /** Usage flows to the supervisor — the budget authority — after EVERY model call, so a
+     *  long tool loop can be terminated mid-flight. */
+    reportUsage(modelRef: string, usage: TokenUsage): void;
+    /** Tell the engine a memory dir is in use — it auto-persists it across runs. */
+    memoryUsed(dir: string): void;
+    /** Broker an MCP OAuth bearer token from the engine (token state never lives here). */
+    mcpToken(serverUrl: string, invalidateToken?: string): Promise<McpTokenResult>;
+    redactor: Redactor;
+    /** Capability context: where the workspace and deployed skills live. */
+    capabilities: ToolSetContext;
+    provider?: ProviderIo;
+}
+/** Execute one agent() leaf call; resolves to final text, or the parsed object in schema mode. */
+export declare function runAgentLeaf(prompt: string, opts: AgentOptions | undefined, io: LeafIo): Promise<unknown>;

package/dist/agent/leaf.js

@@ -0,0 +1,190 @@
+// The agent() leaf: a real agentic loop (SDK SPEC §2.1.1) — streamed model turns with tool
+// use (program-defined ToolDefs + memory file tools + MCP server tools), skills loaded into
+// context, schema output, secret redaction, and usage reporting. Runs IN THE PROGRAM PROCESS
+// (tool `execute` must run in the trusted layer, and MCP connections live where their tools
+// execute); model/provider/key resolution and MCP OAuth token state stay supervisor-side and
+// arrive through `io.resolve` / `io.mcpToken`.
+import { randomUUID } from "node:crypto";
+import { EngineError } from "../errors.js";
+import { chatAnthropic, chatOpenAi } from "./providers.js";
+import { assertUniqueToolNames, buildToolSet, connectMcpServers, } from "./tools.js";
+/** Tool iterations per agent() call before the loop is declared runaway. */
+const MAX_TOOL_ITERATIONS = 25;
+/** Execute one agent() leaf call; resolves to final text, or the parsed object in schema mode. */
+export async function runAgentLeaf(prompt, opts, io) {
+    const base = buildToolSet(opts, io.capabilities);
+    if (base.memoryDir !== null)
+        io.memoryUsed(base.memoryDir);
+    // MCP connections open AFTER sync validation and BEFORE the first model call; a server the
+    // call named must resolve or the leaf fails here, never silently degrading the tool set.
+    const connected = base.mcp.length === 0
+        ? null
+        : await connectMcpServers(base.mcp, {
+            mcpToken: (serverUrl, invalidateToken) => io.mcpToken(serverUrl, invalidateToken),
+            redactor: io.redactor,
+        });
+    try {
+        return await runLeafWithTools(prompt, opts, io, [...base.tools, ...(connected?.tools ?? [])], base.preamble);
+    }
+    finally {
+        // Disconnect on completion AND on error — stdio servers are real child processes.
+        if (connected !== null)
+            await connected.disconnect();
+    }
+}
+async function runLeafWithTools(prompt, opts, io, tools, preamble) {
+    // Re-check across the MERGED set: a namespaced MCP tool can collide with a program tool.
+    assertUniqueToolNames(tools);
+    const resolved = await io.resolve(opts?.model, opts?.provider);
+    // The provider key (and any env-sourced custom headers) are now known to this process —
+    // make sure they can never reach the model.
+    if (resolved.apiKey !== null)
+        io.redactor.add(`api-key:${resolved.provider}`, resolved.apiKey);
+    for (const name of resolved.secretHeaderNames) {
+        const value = resolved.headers[name];
+        if (value !== undefined)
+            io.redactor.add(`header:${name}`, value);
+    }
+    const schemaInstruction = opts?.schema === undefined
+        ? []
+        : [
+            `Respond with ONLY a JSON value matching this JSON Schema — no prose, no code fences:\n${JSON.stringify(opts.schema)}`,
+        ];
+    const firstMessage = [...preamble, prompt, ...schemaInstruction].join("\n\n");
+    const turnId = `turn-${randomUUID()}`;
+    io.startTurn(turnId);
+    const totals = {
+        inputTokens: 0,
+        outputTokens: 0,
+    };
+    const messages = [{ role: "user", text: io.redactor.redact(firstMessage) }];
+    let finalText;
+    try {
+        finalText = await runToolLoop(messages, tools, resolved, io, turnId, totals);
+    }
+    catch (err) {
+        // Redact like the tool path does: a misbehaving provider can echo request headers (the
+        // API key) into an error body, and this message persists in run_events AND — via the
+        // rethrow — in the run's failed row. The secrets invariant covers error paths too.
+        const message = io.redactor.redact(err instanceof Error ? err.message : String(err));
+        const code = err instanceof EngineError ? err.code : "PROVIDER_ERROR";
+        io.emit(turnId, {
+            kind: "turn_ended",
+            ...io.identity,
+            reason: "error",
+            error: { code, message },
+        });
+        throw new EngineError(code, message, err instanceof EngineError && err.hint !== undefined
+            ? io.redactor.redact(err.hint)
+            : undefined);
+    }
+    io.emit(turnId, {
+        kind: "turn_ended",
+        ...io.identity,
+        reason: "complete",
+        usage: { inputTokens: totals.inputTokens, outputTokens: totals.outputTokens },
+    });
+    return opts?.schema === undefined ? finalText : parseSchemaOutput(finalText);
+}
+async function runToolLoop(messages, tools, resolved, io, turnId, totals) {
+    for (let iteration = 1; iteration <= MAX_TOOL_ITERATIONS; iteration++) {
+        const turn = await modelTurn(messages, tools, resolved, io, turnId, iteration);
+        totals.inputTokens += turn.usage.inputTokens ?? 0;
+        totals.outputTokens += turn.usage.outputTokens ?? 0;
+        io.reportUsage(resolved.model, turn.usage);
+        if (!turn.wantsTools || turn.toolCalls.length === 0) {
+            return turn.text;
+        }
+        messages.push({ role: "assistant", text: turn.text, toolCalls: turn.toolCalls });
+        const results = [];
+        for (const call of turn.toolCalls) {
+            results.push(await executeToolCall(call, tools, io, turnId));
+        }
+        messages.push({ role: "tool_results", results });
+    }
+    throw new EngineError("PROGRAM_ERROR", `agent() exceeded ${String(MAX_TOOL_ITERATIONS)} tool iterations without a final answer.`, "The model is looping on tool calls. Tighten the prompt, or split the work across calls.");
+}
+/** One model call with streamed text events (block ids are unique per iteration). */
+async function modelTurn(messages, tools, resolved, io, turnId, iteration) {
+    const blockId = `text-${String(iteration)}`;
+    let blockOpen = false;
+    const providerIo = {
+        ...io.provider,
+        onDelta: (text) => {
+            if (!blockOpen) {
+                blockOpen = true;
+                io.emit(turnId, { kind: "text_start", blockId });
+            }
+            io.emit(turnId, { kind: "text_delta", blockId, text: io.redactor.redact(text) });
+        },
+    };
+    const args = {
+        baseUrl: resolved.baseUrl,
+        apiKey: resolved.apiKey,
+        headers: resolved.headers,
+        model: resolved.model,
+        messages,
+        tools: tools.map(({ name, description, inputSchema }) => ({ name, description, inputSchema })),
+    };
+    const turn = resolved.protocol === "anthropic"
+        ? await chatAnthropic(args, providerIo)
+        : await chatOpenAi(args, providerIo);
+    if (blockOpen)
+        io.emit(turnId, { kind: "text_end", blockId });
+    return turn;
+}
+/** Run one tool call; failures return to the MODEL as error results, they don't fail the run. */
+async function executeToolCall(call, tools, io, turnId) {
+    io.emit(turnId, { kind: "tool_call_start", toolCallId: call.id, toolName: call.name });
+    io.emit(turnId, { kind: "tool_call_input_complete", toolCallId: call.id, input: call.input });
+    const tool = tools.find((t) => t.name === call.name);
+    if (tool === undefined) {
+        // The model invented a tool name — its mistake to recover from, not a run failure.
+        const message = `Unknown tool "${call.name}".`;
+        io.emit(turnId, {
+            kind: "tool_call_error",
+            toolCallId: call.id,
+            error: { code: "VALIDATION", message },
+        });
+        return { id: call.id, content: message, isError: true };
+    }
+    io.emit(turnId, { kind: "tool_call_executing", toolCallId: call.id });
+    try {
+        // Tool results enter model context → redact (MASTER_SPEC §6.2 covers tool results too).
+        const content = io.redactor.redact(await tool.execute(call.input));
+        io.emit(turnId, {
+            kind: "tool_call_result",
+            toolCallId: call.id,
+            result: { humanSummary: summarize(content) },
+        });
+        return { id: call.id, content, isError: false };
+    }
+    catch (err) {
+        const message = io.redactor.redact(err instanceof Error ? err.message : String(err));
+        const code = err instanceof EngineError ? err.code : "PROGRAM_ERROR";
+        io.emit(turnId, { kind: "tool_call_error", toolCallId: call.id, error: { code, message } });
+        return { id: call.id, content: `Tool failed: ${message}`, isError: true };
+    }
+}
+function summarize(content) {
+    const flat = content.replaceAll("\n", " ");
+    return flat.length <= 120 ? flat : `${flat.slice(0, 117)}…`;
+}
+/**
+ * Schema mode: parse the model's final text as JSON; a non-JSON answer fails the run (the
+ * documented contract). Code fences are stripped first — models add them despite instructions.
+ * v0 parity note: like the hosted platform today, the schema drives the PROMPT and the JSON
+ * parse; structural validation against the schema itself is a later, cross-engine decision.
+ */
+function parseSchemaOutput(text) {
+    const stripped = text
+        .trim()
+        .replace(/^```(?:json)?\s*\n?/i, "")
+        .replace(/\n?```\s*$/, "");
+    try {
+        return JSON.parse(stripped);
+    }
+    catch {
+        throw new EngineError("VALIDATION", "agent() was called with a schema but the model's answer was not valid JSON.", `Answer started with: ${JSON.stringify(stripped.slice(0, 80))}`);
+    }
+}

package/dist/agent/providers.d.ts

@@ -0,0 +1,23 @@
+import type { ChatMessage, ChatTurn, ToolSpec } from "./conversation.js";
+export interface ChatArgs {
+    baseUrl: string;
+    apiKey: string | null;
+    /** Extra request headers (custom auth schemes etc.). WIN over computed auth on collision;
+     *  content-type stays engine-owned. */
+    headers?: Record<string, string>;
+    /** Opaque model string, sent verbatim. */
+    model: string;
+    messages: readonly ChatMessage[];
+    tools: readonly ToolSpec[];
+    /** Anthropic requires max_tokens; this default is deliberately generous. */
+    maxTokens?: number;
+}
+/** Injectable effects so adapter logic is unit-testable without a network or real waits. */
+export interface ProviderIo {
+    fetchImpl?: typeof fetch;
+    sleepImpl?: (ms: number) => Promise<void>;
+    /** Streamed text chunks, in order, as they arrive (drives text_delta events). */
+    onDelta?: (text: string) => void;
+}
+export declare function chatAnthropic(args: ChatArgs, io?: ProviderIo): Promise<ChatTurn>;
+export declare function chatOpenAi(args: ChatArgs, io?: ProviderIo): Promise<ChatTurn>;