feat: add optional llm-task JSON-only tool (#1498)

* feat(llm-task): add optional JSON-only LLM task tool

* fix(llm-task): fix invalid package.json

* fix(llm-task): fix invalid plugin manifest JSON

* fix(llm-task): fix index.ts import quoting

* fix(llm-task): load embedded runner from src or bundled dist
This commit is contained in:
Vignesh
2026-01-23 17:18:47 -08:00
committed by GitHub
parent cb06e133ca
commit 95d45c0aa7
6 changed files with 415 additions and 0 deletions

View File

@@ -0,0 +1,86 @@
# LLM Task (plugin)
Adds an **optional** agent tool `llm-task` for running **JSON-only** LLM tasks (drafting, summarizing, classifying) with optional JSON Schema validation.
This is designed to be called from workflow engines (e.g. Lobster via `clawd.invoke --each`) without adding new Clawdbot code per workflow.
## Enable
1) Enable the plugin:
```json
{
"plugins": {
"entries": {
"llm-task": { "enabled": true }
}
}
}
```
2) Allowlist the tool (it is registered with `optional: true`):
```json
{
"agents": {
"list": [
{
"id": "main",
"tools": { "allow": ["llm-task"] }
}
]
}
}
```
## Config (optional)
```json
{
"plugins": {
"entries": {
"llm-task": {
"enabled": true,
"config": {
"defaultProvider": "openai-codex",
"defaultModel": "gpt-5.2",
"allowedModels": ["openai-codex/gpt-5.2"],
"maxTokens": 800,
"timeoutMs": 30000
}
}
}
}
}
```
`allowedModels` is an allowlist of `provider/model` strings. If set, any request outside the list is rejected.
## Tool API
### Parameters
- `prompt` (string, required)
- `input` (any, optional)
- `schema` (object, optional JSON Schema)
- `provider` (string, optional)
- `model` (string, optional)
- `authProfileId` (string, optional)
- `temperature` (number, optional)
- `maxTokens` (number, optional)
- `timeoutMs` (number, optional)
### Output
Returns `details.json` containing the parsed JSON (and validates against `schema` when provided).
## Notes
- The tool is **JSON-only** and instructs the model to output only JSON (no code fences, no commentary).
- Side effects should be handled outside this tool (e.g. approvals in Lobster) before calling tools that send messages/emails.
## Bundled extension note
This extension depends on Clawdbot internal modules (the embedded agent runner). It is intended to ship as a **bundled** Clawdbot extension (like `lobster`) and be enabled via `plugins.entries` + tool allowlists.
It is **not** currently designed to be copied into `~/.clawdbot/extensions` as a standalone plugin directory.

View File

@@ -0,0 +1,21 @@
{
"id": "llm-task",
"name": "LLM Task",
"description": "Generic JSON-only LLM tool for structured tasks callable from workflows.",
"configSchema": {
"type": "object",
"additionalProperties": false,
"properties": {
"defaultProvider": { "type": "string" },
"defaultModel": { "type": "string" },
"defaultAuthProfileId": { "type": "string" },
"allowedModels": {
"type": "array",
"items": { "type": "string" },
"description": "Allowlist of provider/model keys like openai-codex/gpt-5.2."
},
"maxTokens": { "type": "number" },
"timeoutMs": { "type": "number" }
}
}
}

View File

@@ -0,0 +1,5 @@
import { createLlmTaskTool } from "./src/llm-task-tool.js";
export default function (api: any) {
api.registerTool(createLlmTaskTool(api), { optional: true });
}

View File

@@ -0,0 +1,7 @@
{
"name": "@clawdbot/llm-task",
"private": true,
"type": "module",
"main": "index.ts",
"version": "0.0.0"
}

View File

@@ -0,0 +1,96 @@
import { describe, it, expect, vi, beforeEach } from "vitest";
vi.mock("../../../src/agents/pi-embedded-runner.js", () => {
return {
runEmbeddedPiAgent: vi.fn(async () => ({
meta: { startedAt: Date.now() },
payloads: [{ text: "{}" }],
})),
};
});
import { runEmbeddedPiAgent } from "../../../src/agents/pi-embedded-runner.js";
import { createLlmTaskTool } from "./llm-task-tool.js";
function fakeApi(overrides: any = {}) {
return {
id: "llm-task",
name: "llm-task",
source: "test",
config: { agents: { defaults: { workspace: "/tmp", model: { primary: "openai-codex/gpt-5.2" } } } },
pluginConfig: {},
runtime: { version: "test" },
logger: { debug() {}, info() {}, warn() {}, error() {} },
registerTool() {},
...overrides,
};
}
describe("llm-task tool (json-only)", () => {
beforeEach(() => vi.clearAllMocks());
it("returns parsed json", async () => {
(runEmbeddedPiAgent as any).mockResolvedValueOnce({
meta: {},
payloads: [{ text: JSON.stringify({ foo: "bar" }) }],
});
const tool = createLlmTaskTool(fakeApi() as any);
const res = await tool.execute("id", { prompt: "return foo" });
expect((res as any).details.json).toEqual({ foo: "bar" });
});
it("validates schema", async () => {
(runEmbeddedPiAgent as any).mockResolvedValueOnce({
meta: {},
payloads: [{ text: JSON.stringify({ foo: "bar" }) }],
});
const tool = createLlmTaskTool(fakeApi() as any);
const schema = {
type: "object",
properties: { foo: { type: "string" } },
required: ["foo"],
additionalProperties: false,
};
const res = await tool.execute("id", { prompt: "return foo", schema });
expect((res as any).details.json).toEqual({ foo: "bar" });
});
it("throws on invalid json", async () => {
(runEmbeddedPiAgent as any).mockResolvedValueOnce({ meta: {}, payloads: [{ text: "not-json" }] });
const tool = createLlmTaskTool(fakeApi() as any);
await expect(tool.execute("id", { prompt: "x" })).rejects.toThrow(/invalid json/i);
});
it("throws on schema mismatch", async () => {
(runEmbeddedPiAgent as any).mockResolvedValueOnce({
meta: {},
payloads: [{ text: JSON.stringify({ foo: 1 }) }],
});
const tool = createLlmTaskTool(fakeApi() as any);
const schema = { type: "object", properties: { foo: { type: "string" } }, required: ["foo"] };
await expect(tool.execute("id", { prompt: "x", schema })).rejects.toThrow(/match schema/i);
});
it("passes provider/model overrides to embedded runner", async () => {
(runEmbeddedPiAgent as any).mockResolvedValueOnce({
meta: {},
payloads: [{ text: JSON.stringify({ ok: true }) }],
});
const tool = createLlmTaskTool(fakeApi() as any);
await tool.execute("id", { prompt: "x", provider: "anthropic", model: "claude-4-sonnet" });
const call = (runEmbeddedPiAgent as any).mock.calls[0]?.[0];
expect(call.provider).toBe("anthropic");
expect(call.model).toBe("claude-4-sonnet");
});
it("enforces allowedModels", async () => {
(runEmbeddedPiAgent as any).mockResolvedValueOnce({
meta: {},
payloads: [{ text: JSON.stringify({ ok: true }) }],
});
const tool = createLlmTaskTool(fakeApi({ pluginConfig: { allowedModels: ["openai-codex/gpt-5.2"] } }) as any);
await expect(tool.execute("id", { prompt: "x", provider: "anthropic", model: "claude-4-sonnet" })).rejects.toThrow(
/not allowed/i,
);
});
});

View File

@@ -0,0 +1,200 @@
import os from "node:os";
import path from "node:path";
import fs from "node:fs/promises";
import Ajv from "ajv";
import { Type } from "@sinclair/typebox";
// NOTE: This extension is intended to be bundled with Clawdbot.
// When running from source (tests/dev), Clawdbot internals live under src/.
// When running from a built install, internals live under dist/ (no src/ tree).
// So we resolve internal imports dynamically with src-first, dist-fallback.
import type { ClawdbotPluginApi } from "../../../src/plugins/types.js";
type RunEmbeddedPiAgentFn = (params: any) => Promise<any>;
async function loadRunEmbeddedPiAgent(): Promise<RunEmbeddedPiAgentFn> {
// Source checkout (tests/dev)
try {
const mod = await import("../../../src/agents/pi-embedded-runner.js");
if (typeof (mod as any).runEmbeddedPiAgent === "function") return (mod as any).runEmbeddedPiAgent;
} catch {
// ignore
}
// Bundled install (built)
const mod = await import("../../../agents/pi-embedded-runner.js");
if (typeof (mod as any).runEmbeddedPiAgent !== "function") {
throw new Error("Internal error: runEmbeddedPiAgent not available");
}
return (mod as any).runEmbeddedPiAgent;
}
function stripCodeFences(s: string): string {
const trimmed = s.trim();
const m = trimmed.match(/^```(?:json)?s*([sS]*?)s*```$/i);
if (m) return (m[1] ?? "").trim();
return trimmed;
}
function collectText(payloads: Array<{ text?: string; isError?: boolean }> | undefined): string {
const texts = (payloads ?? [])
.filter((p) => !p.isError && typeof p.text === "string")
.map((p) => p.text ?? "");
return texts.join("n").trim();
}
function toModelKey(provider?: string, model?: string): string | undefined {
const p = provider?.trim();
const m = model?.trim();
if (!p || !m) return undefined;
return `${p}/${m}`;
}
type PluginCfg = {
defaultProvider?: string;
defaultModel?: string;
defaultAuthProfileId?: string;
allowedModels?: string[];
maxTokens?: number;
timeoutMs?: number;
};
export function createLlmTaskTool(api: ClawdbotPluginApi) {
return {
name: "llm-task",
description:
"Run a generic JSON-only LLM task and return schema-validated JSON. Designed for orchestration from Lobster workflows via clawd.invoke.",
parameters: Type.Object({
prompt: Type.String({ description: "Task instruction for the LLM." }),
input: Type.Optional(Type.Unknown({ description: "Optional input payload for the task." })),
schema: Type.Optional(Type.Unknown({ description: "Optional JSON Schema to validate the returned JSON." })),
provider: Type.Optional(Type.String({ description: "Provider override (e.g. openai-codex, anthropic)." })),
model: Type.Optional(Type.String({ description: "Model id override." })),
authProfileId: Type.Optional(Type.String({ description: "Auth profile override." })),
temperature: Type.Optional(Type.Number({ description: "Best-effort temperature override." })),
maxTokens: Type.Optional(Type.Number({ description: "Best-effort maxTokens override." })),
timeoutMs: Type.Optional(Type.Number({ description: "Timeout for the LLM run." })),
}),
async execute(_id: string, params: Record<string, unknown>) {
const prompt = String(params.prompt ?? "");
if (!prompt.trim()) throw new Error("prompt required");
const pluginCfg = (api.pluginConfig ?? {}) as PluginCfg;
const primary = api.config?.agents?.defaults?.model?.primary;
const primaryProvider = typeof primary === "string" ? primary.split("/")[0] : undefined;
const primaryModel = typeof primary === "string" ? primary.split("/").slice(1).join("/") : undefined;
const provider =
(typeof params.provider === "string" && params.provider.trim()) ||
(typeof pluginCfg.defaultProvider === "string" && pluginCfg.defaultProvider.trim()) ||
primaryProvider ||
undefined;
const model =
(typeof params.model === "string" && params.model.trim()) ||
(typeof pluginCfg.defaultModel === "string" && pluginCfg.defaultModel.trim()) ||
primaryModel ||
undefined;
const authProfileId =
(typeof (params as any).authProfileId === "string" && (params as any).authProfileId.trim()) ||
(typeof pluginCfg.defaultAuthProfileId === "string" && pluginCfg.defaultAuthProfileId.trim()) ||
undefined;
const modelKey = toModelKey(provider, model);
if (!provider || !model || !modelKey) {
throw new Error(
`provider/model could not be resolved (provider=${String(provider ?? "")}, model=${String(model ?? "")})`,
);
}
const allowed = Array.isArray(pluginCfg.allowedModels) ? pluginCfg.allowedModels : undefined;
if (allowed && allowed.length > 0 && !allowed.includes(modelKey)) {
throw new Error(
`Model not allowed by llm-task plugin config: ${modelKey}. Allowed models: ${allowed.join(", ")}`,
);
}
const timeoutMs =
(typeof params.timeoutMs === "number" && params.timeoutMs > 0 ? params.timeoutMs : undefined) ||
(typeof pluginCfg.timeoutMs === "number" && pluginCfg.timeoutMs > 0 ? pluginCfg.timeoutMs : undefined) ||
30_000;
const streamParams = {
temperature: typeof params.temperature === "number" ? params.temperature : undefined,
maxTokens:
typeof params.maxTokens === "number"
? params.maxTokens
: typeof pluginCfg.maxTokens === "number"
? pluginCfg.maxTokens
: undefined,
};
const input = (params as any).input as unknown;
const system = [
"You are a JSON-only function.",
"Return ONLY a valid JSON value.",
"Do not wrap in markdown fences.",
"Do not include commentary.",
"Do not call tools.",
].join(" ");
const fullPrompt = `${system}nnTASK:n${prompt}nnINPUT_JSON:n${JSON.stringify(input ?? null, null, 2)}n`;
const tmpDir = await fs.mkdtemp(path.join(os.tmpdir(), "clawdbot-llm-task-"));
const sessionId = `llm-task-${Date.now()}`;
const sessionFile = path.join(tmpDir, "session.json");
const runEmbeddedPiAgent = await loadRunEmbeddedPiAgent();
const result = await runEmbeddedPiAgent({
sessionId,
sessionFile,
workspaceDir: api.config?.agents?.defaults?.workspace ?? process.cwd(),
config: api.config,
prompt: fullPrompt,
timeoutMs,
runId: `llm-task-${Date.now()}`,
provider,
model,
authProfileId,
authProfileIdSource: authProfileId ? "user" : "auto",
streamParams,
});
const text = collectText((result as any).payloads);
if (!text) throw new Error("LLM returned empty output");
const raw = stripCodeFences(text);
let parsed: unknown;
try {
parsed = JSON.parse(raw);
} catch {
throw new Error("LLM returned invalid JSON");
}
const schema = (params as any).schema as unknown;
if (schema && typeof schema === "object") {
const ajv = new Ajv({ allErrors: true, strict: false });
const validate = ajv.compile(schema as any);
const ok = validate(parsed);
if (!ok) {
const msg =
validate.errors?.map((e) => `${e.instancePath || "<root>"} ${e.message || "invalid"}`).join("; ") ??
"invalid";
throw new Error(`LLM JSON did not match schema: ${msg}`);
}
}
return {
content: [{ type: "text", text: JSON.stringify(parsed, null, 2) }],
details: { json: parsed, provider, model },
};
},
};
}