fix: resolve CI log errors — token overflow, hallucinated files, misnamed tool, missing profileRegistry (#108)

* fix: file-list recursive excludes build dirs, workspace exploration before planning, code-run name, profileRegistry in graph adapter

Agent-Logs-Url: https://github.com/huberp/agentloop/sessions/3b124bf9-fd1e-4d37-8483-0cbef986cab9

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

* fix: add CJS shims for jsdom@29 ESM-only deps, fix mock sequences for planNode workspace exploration

Agent-Logs-Url: https://github.com/huberp/agentloop/sessions/f9216c8d-20db-4b33-9e63-2b37f76b0ff8

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

* chore: improve shim comments — fix spelling, document selector limitations

Agent-Logs-Url: https://github.com/huberp/agentloop/sessions/f9216c8d-20db-4b33-9e63-2b37f76b0ff8

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

package.json

@@ -81,7 +81,10 @@
     "moduleNameMapper": {
       "^@modelcontextprotocol/sdk/(.*)$": "<rootDir>/node_modules/@modelcontextprotocol/sdk/dist/cjs/$1",
       "^uuid$": "<rootDir>/node_modules/uuid/dist/index.js",
-      "^@exodus/bytes/(.*)$": "<rootDir>/tests/jest-shims/exodus-bytes-encoding.cjs"
+      "^@exodus/bytes/(.*)$": "<rootDir>/tests/jest-shims/exodus-bytes-encoding.cjs",
+      "^@asamuzakjp/css-color$": "<rootDir>/tests/jest-shims/asamuzakjp-css-color.cjs",
+      "^@asamuzakjp/dom-selector$": "<rootDir>/tests/jest-shims/asamuzakjp-dom-selector.cjs",
+      "^parse5$": "<rootDir>/node_modules/parse5/dist/cjs/index.js"
     }
   }
 }

src/__tests__/langgraph.test.ts

@@ -601,9 +601,12 @@ describe("invokeGraph — integration with mock LLM", () => {
         { type: "step", description: "do thing 2", toolsNeeded: [], estimatedComplexity: "low" },
       ],
     });
+    const workspaceCtxJson = JSON.stringify({
+      workspaceInfo: { language: "node", framework: "none", packageManager: "npm", hasTests: true, testCommand: "", lintCommand: "", buildCommand: "", entryPoints: [], gitInitialized: true },
+    });
 
-    // Mock: planner returns the plan, then each step returns "done"
-    const llm = makeMockLlm(planJson, "step 1 done", "step 2 done");
+    // Mock: explorer returns workspace context, planner returns the plan, then each step returns "done"
+    const llm = makeMockLlm(workspaceCtxJson, planJson, "step 1 done", "step 2 done");
     const registry = new ToolRegistry();
 
     const events: GraphEvent[] = [];
@@ -635,8 +638,12 @@ describe("invokeGraph — integration with mock LLM", () => {
         },
       ],
     });
+    const workspaceCtxJson = JSON.stringify({
+      workspaceInfo: { language: "node", framework: "none", packageManager: "npm", hasTests: true, testCommand: "", lintCommand: "", buildCommand: "", entryPoints: [], gitInitialized: true },
+    });
 
-    const llm = makeMockLlm(planJson, "fast wins!", "slow result");
+    // Explorer → workspace context; planner → plan; steps → results
+    const llm = makeMockLlm(workspaceCtxJson, planJson, "fast wins!", "slow result");
     const registry = new ToolRegistry();
 
     // maxConcurrency=1 forces sequential execution: first branch completes,
@@ -658,12 +665,19 @@ describe("invokeGraph — integration with mock LLM", () => {
         { type: "step", description: "will fail", toolsNeeded: [], estimatedComplexity: "low" },
       ],
     });
+    const workspaceCtxJson = JSON.stringify({
+      workspaceInfo: { language: "node", framework: "none", packageManager: "npm", hasTests: true, testCommand: "", lintCommand: "", buildCommand: "", entryPoints: [], gitInitialized: true },
+    });
 
-    // Planner returns the plan; step LLM throws
+    // Explorer returns workspace context (call 1); planner returns plan (call 2); step throws (call 3+)
     let callCount = 0;
     const invoke = jest.fn().mockImplementation(() => {
       callCount++;
       if (callCount === 1) {
+        // Project-explorer call: return workspace context
+        return Promise.resolve({ content: workspaceCtxJson, tool_calls: [] });
+      }
+      if (callCount === 2) {
         // Planner call: return the plan
         return Promise.resolve({ content: planJson, tool_calls: [] });
       }

src/__tests__/start-oneshot.test.ts

@@ -544,13 +544,18 @@ describe("start-oneshot: getActiveExecutor", () => {
     const original = config.appConfig.orchestrator;
     (config.appConfig as any).orchestrator = "langgraph";
 
-    // Override mock LLM to return a valid blocks plan then a step response
+    // Override mock LLM to return workspace context (explorer), then a valid
+    // blocks plan (planner), then a step response.
+    const workspaceCtxJson = JSON.stringify({
+      workspaceInfo: { language: "node", framework: "none", packageManager: "npm", hasTests: true, testCommand: "", lintCommand: "", buildCommand: "", entryPoints: [], gitInitialized: true },
+    });
     const planJson = JSON.stringify({
       version: "2.0",
       goal: "say hello",
       blocks: [{ type: "step", description: "greet", toolsNeeded: [], estimatedComplexity: "low" }],
     });
     mockLlmInvoke
+      .mockResolvedValueOnce({ content: workspaceCtxJson, tool_calls: [] })
       .mockResolvedValueOnce({ content: planJson, tool_calls: [] })
       .mockResolvedValueOnce({ content: "Hello!", tool_calls: [] });
 
@@ -571,13 +576,18 @@ describe("start-oneshot: getActiveExecutor", () => {
     const original = config.appConfig.orchestrator;
     (config.appConfig as any).orchestrator = "langgraph";
 
-    // Override mock LLM to return a valid blocks plan then a step response
+    // Override mock LLM to return workspace context (explorer), then a valid
+    // blocks plan (planner), then a step response.
+    const workspaceCtxJson = JSON.stringify({
+      workspaceInfo: { language: "node", framework: "none", packageManager: "npm", hasTests: true, testCommand: "", lintCommand: "", buildCommand: "", entryPoints: [], gitInitialized: true },
+    });
     const planJson = JSON.stringify({
       version: "2.0",
       goal: "say hello",
       blocks: [{ type: "step", description: "greet", toolsNeeded: [], estimatedComplexity: "low" }],
     });
     mockLlmInvoke
+      .mockResolvedValueOnce({ content: workspaceCtxJson, tool_calls: [] })
       .mockResolvedValueOnce({ content: planJson, tool_calls: [] })
       .mockResolvedValueOnce({ content: "Hello!", tool_calls: [] });
 

src/agents/builtin/coder.agent.json

@@ -5,7 +5,7 @@
   "model": "gpt-4o",
   "temperature": 0.2,
   "skills": ["typescript-expert"],
-  "tools": ["file-read", "file-write", "file-edit", "file-list", "file-delete", "code-run", "code-search", "shell", "calculate"],
+  "tools": ["file-read", "file-write", "file-edit", "file-list", "file-delete", "code_run", "code-search", "shell", "calculate"],
   "maxIterations": 20,
   "promptTemplate": "system",
   "constraints": {

src/index.ts

@@ -470,7 +470,7 @@ function createGraphExecutorAdapter() {
       const result = await graphExecutor.invoke(
         input,
         { sharedContext: conversationHistory ? { conversationHistory } : {} },
-        { registry: toolRegistry },
+        { registry: toolRegistry, profileRegistry: agentProfileRegistry },
       );
       await chatHistory.addMessage(new AIMessage(result.output));
       return { output: result.output };
@@ -482,7 +482,7 @@ function createGraphExecutorAdapter() {
       const result = await graphExecutor.invoke(
         input,
         { sharedContext: conversationHistory ? { conversationHistory } : {} },
-        { registry: toolRegistry },
+        { registry: toolRegistry, profileRegistry: agentProfileRegistry },
       );
       await chatHistory.addMessage(new AIMessage(result.output));
       yield result.output;

src/langgraph/graph.ts

@@ -22,6 +22,7 @@ import { logger } from "../logger";
 import { ToolRegistry } from "../tools/registry";
 import { runSubagent } from "../subagents/runner";
 import type { AgentProfileRegistry } from "../agents/registry";
+import { exploreWorkspace } from "../agents/project-explorer";
 import { validateBlocksPlan, compileBlocksPlanToDag } from "./compiler";
 import { buildRuntimeContextBody } from "../prompts/context";
 import {
@@ -154,11 +155,29 @@ export function buildGraphNodes(deps: GraphDeps, progressCb?: (evt: GraphEvent)
   async function planNode(state: GraphState): Promise<Partial<GraphState>> {
     logger.info({ request: state.request }, "Graph: generating blocks plan");
     const availableTools = deps.registry.list().map((t) => t.name);
+
+    // Explore the workspace so the planner knows what files/build-systems exist.
+    // This prevents the planner from hallucinating file paths that don't exist.
+    let workspaceCtxStr = "";
+    try {
+      const workspaceCtx = await exploreWorkspace({ registry: deps.registry, llm: deps.llm });
+      workspaceCtxStr = JSON.stringify(workspaceCtx, null, 2);
+      logger.info({ workspaceCtx }, "Graph: workspace exploration complete");
+    } catch (err) {
+      logger.warn(
+        { error: (err as Error).message },
+        "Graph: workspace exploration failed; proceeding without workspace context",
+      );
+    }
+
     const parts: string[] = [
       `Task: ${state.request}`,
       `Available tools: ${availableTools.join(", ") || "(none)"}`,
       buildRuntimeContextBody(),
     ];
+    if (workspaceCtxStr) {
+      parts.push(`Workspace context (use this to understand the actual repo structure before planning steps):\n${workspaceCtxStr}`);
+    }
     const convHistory = state.sharedContext.conversationHistory as string | undefined;
     if (convHistory) {
       parts.push(`Previous conversation context:\n${convHistory}`);

src/tools/file-list.ts

@@ -5,10 +5,38 @@ import type { ToolDefinition } from "./registry";
 import { resolveSafe, globToRegExp } from "./file-utils";
 import { appConfig } from "../config";
 
+/**
+ * Directories that are always skipped during recursive traversal.
+ * These are large build-artifact / dependency trees that have no value for
+ * AI-driven code exploration and can easily produce millions of tokens.
+ */
+const DEFAULT_EXCLUDE_DIRS = new Set([
+  "node_modules",
+  ".git",
+  "dist",
+  "build",
+  "out",
+  "target",       // Rust / Maven
+  ".venv",        // Python virtualenv
+  "venv",
+  "__pycache__",
+  ".cache",
+  ".tox",
+  "coverage",
+  ".nyc_output",
+]);
+
 const schema = z.object({
   path: z.string().default(".").describe("Directory path relative to the workspace root (default: workspace root)"),
   glob: z.string().optional().describe("Optional glob pattern to filter entries (e.g. '*.ts', '**/*.json')"),
   recursive: z.boolean().optional().default(false).describe("List entries recursively (default: false)"),
+  exclude: z
+    .array(z.string())
+    .optional()
+    .describe(
+      "Additional directory names to exclude from recursive traversal " +
+        "(node_modules, .git, dist, build, and other common build-artifact directories are always excluded)"
+    ),
 });
 
 /** A single directory entry returned by file-list. */
@@ -22,8 +50,13 @@ interface FileEntry {
 
 /**
  * Recursively collect entries under `dir`, returning paths relative to `baseDir`.
+ * Directories whose base name appears in `excludeDirs` are silently skipped.
  */
-async function collectEntries(dir: string, baseDir: string): Promise<FileEntry[]> {
+async function collectEntries(
+  dir: string,
+  baseDir: string,
+  excludeDirs: Set<string>
+): Promise<FileEntry[]> {
   const entries = await fs.readdir(dir, { withFileTypes: true });
   const results: FileEntry[] = [];
 
@@ -32,9 +65,11 @@ async function collectEntries(dir: string, baseDir: string): Promise<FileEntry[]
     const relativePath = path.relative(baseDir, fullPath);
 
     if (entry.isDirectory()) {
+      // Skip excluded directories (e.g. node_modules, .git, dist)
+      if (excludeDirs.has(entry.name)) continue;
       results.push({ path: relativePath, type: "directory" });
       // Recurse into sub-directory.
-      results.push(...(await collectEntries(fullPath, baseDir)));
+      results.push(...(await collectEntries(fullPath, baseDir, excludeDirs)));
     } else if (entry.isFile()) {
       const stat = await fs.stat(fullPath);
       results.push({ path: relativePath, type: "file", sizeBytes: stat.size });
@@ -55,17 +90,21 @@ export const toolDefinition: ToolDefinition = {
     path: dirPath = ".",
     glob,
     recursive = false,
+    exclude = [],
   }: {
     path?: string;
     glob?: string;
     recursive?: boolean;
+    exclude?: string[];
   }): Promise<string> => {
     const resolved = resolveSafe(appConfig.workspaceRoot, dirPath);
 
     let entries: FileEntry[];
 
     if (recursive) {
-      entries = await collectEntries(resolved, resolved);
+      // Merge caller-supplied exclusions with the built-in defaults
+      const excludeDirs = new Set([...DEFAULT_EXCLUDE_DIRS, ...exclude]);
+      entries = await collectEntries(resolved, resolved, excludeDirs);
     } else {
       const rawEntries = await fs.readdir(resolved, { withFileTypes: true });
       entries = await Promise.all(

tests/jest-shims/asamuzakjp-css-color.cjs

@@ -0,0 +1,42 @@
+"use strict";
+// CJS shim for @asamuzakjp/css-color (ESM-only in npm package).
+// jsdom@29 imports this module for CSS color resolution.  The agent-loop
+// tests do not exercise CSS rendering, so stub implementations that pass
+// through / return safe defaults are sufficient.
+//
+// NOTE: This shim returns pass-through values only. Full CSS color parsing
+// is not implemented; tests that require accurate color computation should
+// not rely on this shim.
+
+/**
+ * Resolve a CSS colour value to a canonical form.
+ * Returns the input unchanged when resolution is not needed.
+ */
+function resolve(value, opt) {
+  if (typeof value === "string") return value;
+  return "";
+}
+
+/**
+ * Convert a colour between spaces.
+ * Returns null (indicating "not resolved") so jsdom falls back gracefully.
+ */
+function convert(value, opt) {
+  return null;
+}
+
+/** Utility helpers exposed by the real package. Stubbed as no-ops. */
+const utils = {
+  cssCalc: (value) => value,
+  cssVar: (value) => value,
+  extractDashedIdent: () => [],
+  isAbsoluteFontSize: () => false,
+  isAbsoluteSizeOrLength: () => false,
+  isColor: (value) => typeof value === "string",
+  isGradient: () => false,
+  resolveGradient: (value) => value,
+  resolveLengthInPixels: () => 0,
+  splitValue: (value) => (typeof value === "string" ? [value] : []),
+};
+
+module.exports = { resolve, convert, utils };