fix(C-2): inject original request into step prompts; require concrete values in planner descriptions (#118)

* Initial plan

* fix(C-2): inject original request into step prompts and strengthen planner instructions

Agent-Logs-Url: https://github.com/huberp/agentloop/sessions/c6821f70-b870-4117-a6d4-06d33377c44f

Co-authored-by: huberp <4027454+huberp@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: huberp <4027454+huberp@users.noreply.github.com>

Author: Copilot

src/__tests__/langgraph.test.ts

@@ -22,9 +22,11 @@ import {
   isDeadlocked,
 } from "../langgraph/scheduler";
 import { buildGraphNodes, invokeGraph } from "../langgraph/graph";
+import { runPlannedStep } from "../langgraph/step-runner";
 import type {
   BlocksPlan,
   CompiledPlan,
+  CompiledPlanNode,
   NodeRecord,
   GraphState,
   GraphEvent,
@@ -728,3 +730,129 @@ describe("finalize node", () => {
     expect(result.output).toContain("Something broke");
   });
 });
+
+// ─────────────────────────────────────────────────────────────────────────────
+// (11) runPlannedStep — originalRequest grounding
+// ─────────────────────────────────────────────────────────────────────────────
+
+describe("runPlannedStep — original request grounding", () => {
+  /** Minimal CompiledPlanNode for unit tests. */
+  function makeNode(overrides: Partial<CompiledPlanNode> = {}): CompiledPlanNode {
+    return {
+      id: "s1",
+      description: "Clone the repository locally",
+      dependsOn: [],
+      toolsNeeded: [],
+      estimatedComplexity: "low",
+      resources: [],
+      ...overrides,
+    };
+  }
+
+  it("includes originalRequest in the step system prompt when provided", async () => {
+    let capturedSystemPrompt = "";
+
+    const invoke = jest.fn().mockImplementation(() => {
+      return Promise.resolve({ content: "done", tool_calls: [] });
+    });
+    const bindTools = jest.fn().mockImplementation((_tools: unknown, opts?: { tool_choice?: string }) => {
+      // Capture the system prompt from whatever call is made
+      return {
+        invoke: jest.fn().mockImplementation((messages: unknown[]) => {
+          if (Array.isArray(messages)) {
+            const systemMsg = (messages as Array<{ _getType?: () => string; content?: string }>)
+              .find((m) => m._getType?.() === "system");
+            if (systemMsg?.content) capturedSystemPrompt = systemMsg.content as string;
+          }
+          return Promise.resolve({ content: "done", tool_calls: [] });
+        }),
+      };
+    });
+
+    const llm = { invoke, bindTools } as unknown as BaseChatModel;
+    const registry = new ToolRegistry();
+    const node = makeNode();
+
+    await runPlannedStep(node, {
+      registry,
+      llm,
+      originalRequest: "add Anthropic models to github repo huberp/agentloop",
+    });
+
+    // The system prompt passed to runSubagent must contain the original request
+    expect(capturedSystemPrompt).toContain("add Anthropic models to github repo huberp/agentloop");
+    expect(capturedSystemPrompt).toContain("Original user request (for context):");
+  });
+
+  it("omits the original-request line when originalRequest is not provided", async () => {
+    let capturedSystemPrompt = "";
+
+    const bindTools = jest.fn().mockImplementation(() => ({
+      invoke: jest.fn().mockImplementation((messages: unknown[]) => {
+        if (Array.isArray(messages)) {
+          const systemMsg = (messages as Array<{ _getType?: () => string; content?: string }>)
+            .find((m) => m._getType?.() === "system");
+          if (systemMsg?.content) capturedSystemPrompt = systemMsg.content as string;
+        }
+        return Promise.resolve({ content: "done", tool_calls: [] });
+      }),
+    }));
+
+    const llm = {
+      invoke: jest.fn().mockResolvedValue({ content: "done", tool_calls: [] }),
+      bindTools,
+    } as unknown as BaseChatModel;
+    const registry = new ToolRegistry();
+    const node = makeNode();
+
+    await runPlannedStep(node, { registry, llm });
+
+    expect(capturedSystemPrompt).not.toContain("Original user request (for context):");
+  });
+
+  it("propagates request from state.request via invokeGraph", async () => {
+    const capturedSystemPrompts: string[] = [];
+
+    const planJson = JSON.stringify({
+      version: "2.0",
+      goal: "clone test",
+      blocks: [
+        { type: "step", description: "Clone the forked repository locally to the workspace", toolsNeeded: [], estimatedComplexity: "low" },
+      ],
+    });
+    const workspaceCtxJson = JSON.stringify({
+      workspaceInfo: { language: "node", framework: "none", packageManager: "npm", hasTests: true, testCommand: "", lintCommand: "", buildCommand: "", entryPoints: [], gitInitialized: true },
+    });
+
+    // Capture all system prompts seen during execution
+    let callCount = 0;
+    const invoke = jest.fn().mockImplementation((messages: unknown[]) => {
+      callCount++;
+      if (Array.isArray(messages)) {
+        const systemMsg = (messages as Array<{ _getType?: () => string; content?: string }>)
+          .find((m) => m._getType?.() === "system");
+        if (systemMsg?.content) capturedSystemPrompts.push(systemMsg.content as string);
+      }
+      if (callCount === 1) return Promise.resolve({ content: workspaceCtxJson, tool_calls: [] });
+      if (callCount === 2) return Promise.resolve({ content: planJson, tool_calls: [] });
+      return Promise.resolve({ content: "cloned successfully", tool_calls: [] });
+    });
+    const llm = {
+      invoke,
+      bindTools: jest.fn().mockImplementation(() => ({ invoke })),
+    } as unknown as BaseChatModel;
+
+    const registry = new ToolRegistry();
+    await invokeGraph(
+      "add Anthropic models to github repo huberp/agentloop",
+      { registry, llm },
+    );
+
+    // At least one system prompt (from the step subagent) must contain the original request
+    const stepPrompts = capturedSystemPrompts.filter((p) =>
+      p.includes("executing one step of a larger plan"),
+    );
+    expect(stepPrompts.length).toBeGreaterThan(0);
+    expect(stepPrompts[0]).toContain("add Anthropic models to github repo huberp/agentloop");
+  }, 30000);
+});

src/langgraph/graph.ts

@@ -116,7 +116,8 @@ const BLOCKS_PLANNER_SYSTEM =
   `- Set join to "any" when only the first successful branch matters.\n` +
   `- Mark resources: ["network"] for steps using web search/fetch.\n` +
   `- Mark resources: ["file:WRITE:<path>"] for steps writing to a specific file.\n` +
-  `- Produce at least one block.`;
+  `- Produce at least one block.\n` +
+  `- Include all concrete values (URLs, repository names, file paths, version numbers, package names) needed to execute the step directly in the step description itself.`;
 
 // ─────────────────────────────────────────────────────────────────────────────
 // Dependencies injected when building the graph
@@ -288,6 +289,7 @@ export function buildGraphNodes(deps: GraphDeps, progressCb?: (evt: GraphEvent)
           llm: deps.llm,
           profileRegistry: deps.profileRegistry,
           sharedContext: state.sharedContext,
+          originalRequest: state.request,
         });
         return { nodeId, ...result };
       }),

src/langgraph/step-runner.ts

@@ -45,6 +45,8 @@ export interface StepRunnerDeps {
   profileRegistry?: AgentProfileRegistry;
   /** Shared context from the graph state (conversation history, prior step outputs). */
   sharedContext?: Record<string, unknown>;
+  /** The original user request; injected into every step prompt to prevent hallucination. */
+  originalRequest?: string;
 }
 
 /**
@@ -102,6 +104,7 @@ export async function runPlannedStep(
 
   const stepSystemPrompt =
     `You are an AI agent executing one step of a larger plan.\n` +
+    (deps.originalRequest ? `Original user request (for context): ${deps.originalRequest}\n` : ``) +
     `Step: ${node.description}\n` +
     `${toolList}\n` +
     `Instructions:\n` +