fix(miner): derive sdk changed files from git

packages/gittensory-engine/src/miner/agent-sdk-driver.ts

@@ -8,6 +8,9 @@
 // matcher, #2343's stated attachment point) and this driver forwards them verbatim onto the `query()` options, so
 // house-rule enforcement can intercept every tool call before execution without this module knowing the rules.
 
+import { execFile } from "node:child_process";
+import { promisify } from "node:util";
+
 import { redactSecrets } from "../subprocess-env.js";
 import type {
   CodingAgentDriver,
@@ -40,8 +43,7 @@ export type AgentSdkQueryFn = (input: {
   options: AgentSdkQueryOptions;
 }) => AsyncIterable<Record<string, unknown>>;
 
-/** Tool names whose successful use means a file in the working directory changed. */
-const FILE_EDIT_TOOL_NAMES = new Set(["Edit", "Write", "NotebookEdit"]);
+const execFileAsync = promisify(execFile);
 
 /** Ceiling for any redacted free text surfaced on the result (error detail, summary) — one named place. */
 const MAX_REDACTED_TEXT_LENGTH = 500;
@@ -66,12 +68,30 @@ export type CreateAgentSdkDriverOptions = {
   query?: AgentSdkQueryFn | undefined;
   /** Forwarded verbatim to the SDK session — the #2343 `PreToolUse` interception point. */
   hooks?: AgentSdkHooks | undefined;
+  /** Injected changed-file enumerator; defaults to git diff over the worktree. */
+  listChangedFiles?: ((cwd: string) => Promise<string[]>) | undefined;
 };
 
 function asRecord(value: unknown): Record<string, unknown> | null {
   return typeof value === "object" && value !== null ? (value as Record<string, unknown>) : null;
 }
 
+async function listWorktreeChangedFiles(cwd: string): Promise<string[]> {
+  const [tracked, untracked] = await Promise.all([
+    execFileAsync("git", ["-C", cwd, "diff", "--name-only", "HEAD", "--"]),
+    execFileAsync("git", ["-C", cwd, "ls-files", "--others", "--exclude-standard"]),
+  ]);
+  return Array.from(
+    new Set(
+      [tracked.stdout, untracked.stdout]
+        .join("\n")
+        .split(/\r?\n/)
+        .map((file) => file.trim())
+        .filter(Boolean),
+    ),
+  );
+}
+
 /** Fold one assistant message's content blocks into the transcript/changed-file accumulators. */
 function foldAssistantMessage(
   message: Record<string, unknown>,
@@ -87,9 +107,7 @@ function foldAssistantMessage(
       transcript.push(block.text);
     } else if (block.type === "tool_use" && typeof block.name === "string") {
       const filePath = asRecord(block.input)?.file_path;
-      if (FILE_EDIT_TOOL_NAMES.has(block.name) && typeof filePath === "string") {
-        changedFiles.add(filePath);
-      }
+      if (typeof filePath === "string") changedFiles.add(filePath);
     }
   }
 }
@@ -105,6 +123,7 @@ export function createAgentSdkCodingAgentDriver(
   options: CreateAgentSdkDriverOptions = {},
 ): CodingAgentDriver {
   const query = options.query ?? defaultQuery;
+  const listChangedFiles = options.listChangedFiles ?? listWorktreeChangedFiles;
 
   return {
     async run(task: CodingAgentDriverTask): Promise<CodingAgentDriverResult> {
@@ -179,10 +198,27 @@ export function createAgentSdkCodingAgentDriver(
         };
       }
 
+      let worktreeChangedFiles: string[];
+      try {
+        worktreeChangedFiles = await listChangedFiles(task.workingDirectory);
+      } catch (error) {
+        const detail = redactSecrets(error instanceof Error ? error.message : String(error)).slice(0, MAX_REDACTED_TEXT_LENGTH);
+        return {
+          ok: false,
+          changedFiles: [],
+          summary: "agent sdk changed-file enumeration failed",
+          transcript,
+          turnsUsed,
+          costUsd,
+          error: `agent_sdk_changed_files_unavailable: ${detail}`,
+        };
+      }
+
+      const allChangedFiles = Array.from(new Set([...changedFiles, ...worktreeChangedFiles]));
       return {
         ok: true,
-        changedFiles: [...changedFiles],
-        summary: resultText.slice(0, MAX_REDACTED_TEXT_LENGTH) || `coding agent completed with ${changedFiles.size} changed file(s)`,
+        changedFiles: allChangedFiles,
+        summary: resultText.slice(0, MAX_REDACTED_TEXT_LENGTH) || `coding agent completed with ${allChangedFiles.length} changed file(s)`,
         transcript,
         turnsUsed,
         costUsd,

packages/gittensory-engine/src/miner/driver-factory.ts

@@ -116,6 +116,10 @@ export type CreateCodingAgentDriverOptions = {
   query?: AgentSdkQueryFn | undefined;
   /** Forwarded to the `agent-sdk` provider's session (#2343's PreToolUse interception point). */
   hooks?: AgentSdkHooks | undefined;
+  /** Optional injected changed-file enumerator for the `agent-sdk` provider (defaults to a real `git diff`/
+   *  `git ls-files` read over the task's working directory — a test harness pointing at a non-git fake path
+   *  should inject one, same seam as `query`). */
+  listChangedFiles?: ((cwd: string) => Promise<string[]>) | undefined;
   /** Known secret values the CLI providers strip from surfaced output, on top of the token-shape patterns. */
   knownSecrets?: readonly string[] | undefined;
 };
@@ -190,6 +194,7 @@ export function createCodingAgentDriver(options: CreateCodingAgentDriverOptions)
       return createAgentSdkCodingAgentDriver({
         ...(options.query !== undefined ? { query: options.query } : {}),
         ...(options.hooks !== undefined ? { hooks: options.hooks } : {}),
+        ...(options.listChangedFiles !== undefined ? { listChangedFiles: options.listChangedFiles } : {}),
       });
     /* v8 ignore next 2 -- isConfiguredCodingAgentDriver already rejects unknown names before this switch. */
     default:
@@ -209,6 +214,8 @@ export type RunCodingAgentAttemptOptions = {
   spawn?: CliSubprocessSpawnFn | undefined;
   query?: AgentSdkQueryFn | undefined;
   hooks?: AgentSdkHooks | undefined;
+  /** Optional injected changed-file enumerator for the `agent-sdk` provider (see `CreateCodingAgentDriverOptions`). */
+  listChangedFiles?: ((cwd: string) => Promise<string[]>) | undefined;
   knownSecrets?: readonly string[] | undefined;
   /** When supplied, the driver result is run through the lint guard (#4276) before being returned, so a
    *  live coding-agent edit that fails its own package's typecheck/node --check never reads as `ok: true`. */
@@ -227,6 +234,7 @@ function resolveDriverForAttempt(options: RunCodingAgentAttemptOptions, mode: Co
     spawn: options.spawn,
     query: options.query,
     hooks: options.hooks,
+    listChangedFiles: options.listChangedFiles,
     knownSecrets: options.knownSecrets,
   });
 }

packages/gittensory-engine/test/agent-sdk-driver.test.ts

@@ -3,6 +3,7 @@ import assert from "node:assert/strict";
 import {
   createAgentSdkCodingAgentDriver,
   type AgentSdkQueryFn,
+  type CreateAgentSdkDriverOptions,
   type CodingAgentDriverTask,
 } from "../dist/index.js";
 
@@ -35,10 +36,14 @@ function queryYielding(
   };
 }
 
+function driverWith(options: CreateAgentSdkDriverOptions) {
+  return createAgentSdkCodingAgentDriver({ listChangedFiles: async () => [], ...options });
+}
+
 test("success: session options, tool-use changed-file tracking, transcript, turn count", async () => {
   const captured: { input?: Parameters<AgentSdkQueryFn>[0] } = {};
   const hooks = { PreToolUse: [{ hooks: ["policy-callback"] }] };
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: queryYielding(
       [
         assistantMessage({ type: "text", text: "editing now" }),
@@ -74,8 +79,62 @@ test("success: session options, tool-use changed-file tracking, transcript, turn
   assert.equal(captured.input!.options.hooks, hooks);
 });
 
+test("success derives changed files from the worktree after untracked mutating tools", async () => {
+  const driver = driverWith({
+    query: queryYielding([
+      assistantMessage({ type: "tool_use", name: "Bash", input: { command: "node mutate.js" } }),
+      { type: "result", subtype: "success", is_error: false, num_turns: 2, result: "mutated" },
+    ]),
+    listChangedFiles: async (cwd) => {
+      assert.equal(cwd, task.workingDirectory);
+      return ["packages/gittensory-engine/src/vulnerable.ts"];
+    },
+  });
+
+  const result = await driver.run(task);
+
+  assert.equal(result.ok, true);
+  assert.deepEqual(result.changedFiles, ["packages/gittensory-engine/src/vulnerable.ts"]);
+});
+
+test("success fails closed when changed-file enumeration is unavailable, but still reports the real dollar cost", async () => {
+  const driver = driverWith({
+    query: queryYielding([
+      { type: "result", subtype: "success", is_error: false, num_turns: 2, result: "done", total_cost_usd: 0.0042 },
+    ]),
+    listChangedFiles: async () => {
+      throw new Error("not a git worktree");
+    },
+  });
+
+  const result = await driver.run(task);
+
+  assert.equal(result.ok, false);
+  assert.deepEqual(result.changedFiles, []);
+  assert.match(result.error!, /agent_sdk_changed_files_unavailable: not a git worktree/);
+  // The SDK session ran and was billed before enumeration ever failed -- budgetSpent must not silently
+  // undercount this path just because the changed-files step failed afterward.
+  assert.equal(result.costUsd, 0.0042);
+});
+
+test("success stringifies a non-Error changed-file enumeration failure", async () => {
+  const driver = driverWith({
+    query: queryYielding([
+      { type: "result", subtype: "success", is_error: false, num_turns: 2, result: "done" },
+    ]),
+    listChangedFiles: async () => {
+      throw "git unavailable";
+    },
+  });
+
+  const result = await driver.run(task);
+
+  assert.equal(result.ok, false);
+  assert.equal(result.error, "agent_sdk_changed_files_unavailable: git unavailable");
+});
+
 test("non-success result subtype maps to a structured failure with the subtype as the error", async () => {
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: queryYielding([
       assistantMessage({ type: "tool_use", name: "Edit", input: { file_path: "src/a.ts" } }),
       { type: "result", subtype: "error_max_turns", is_error: true, num_turns: 6 },
@@ -90,7 +149,7 @@ test("non-success result subtype maps to a structured failure with the subtype a
 });
 
 test("a success-subtype result that still flags is_error is treated as a failure", async () => {
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: queryYielding([
       { type: "result", subtype: "success", is_error: true, num_turns: 1, result: "refused" },
     ]),
@@ -101,7 +160,7 @@ test("a success-subtype result that still flags is_error is treated as a failure
 });
 
 test("stream ending without a result frame is a protocol failure, not a silent success", async () => {
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: queryYielding([assistantMessage({ type: "text", text: "started..." })]),
   });
   const result = await driver.run(task);
@@ -111,7 +170,7 @@ test("stream ending without a result frame is a protocol failure, not a silent s
 });
 
 test("a throw mid-stream returns a redacted structured failure and never propagates", async () => {
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: () =>
       (async function* (): AsyncGenerator<Record<string, unknown>> {
         yield assistantMessage({ type: "text", text: "before the crash" });
@@ -127,7 +186,7 @@ test("a throw mid-stream returns a redacted structured failure and never propaga
 });
 
 test("secret shapes in the result text are redacted from summary and transcript", async () => {
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: queryYielding([
       {
         type: "result",
@@ -146,7 +205,7 @@ test("secret shapes in the result text are redacted from summary and transcript"
 });
 
 test("malformed frames (no content array, non-object blocks, missing file_path) are skipped defensively", async () => {
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: queryYielding([
       { type: "assistant" },
       { type: "assistant", message: { content: "not-an-array" } },
@@ -167,7 +226,7 @@ test("malformed frames (no content array, non-object blocks, missing file_path)
 });
 
 test("names a result frame with no usable subtype 'unknown'", async () => {
-  const driver = createAgentSdkCodingAgentDriver({
+  const driver = driverWith({
     query: queryYielding([{ type: "result", is_error: true }]),
   });
   const result = await driver.run(task);

packages/gittensory-miner/lib/coding-agent-construction.d.ts

@@ -6,6 +6,7 @@ export type ConstructProductionCodingAgentDriverOptions = {
   spawn?: CliSubprocessSpawnFn;
   query?: AgentSdkQueryFn;
   hooks?: AgentSdkHooks;
+  listChangedFiles?: (cwd: string) => Promise<string[]>;
   houseRulesConfig?: unknown;
   houseRulesOptions?: unknown;
 };

packages/gittensory-miner/lib/coding-agent-construction.js

@@ -76,6 +76,7 @@ export function createRealCliSubprocessSpawn() {
  *   spawn?: import("@jsonbored/gittensory-engine").CliSubprocessSpawnFn,
  *   query?: import("@jsonbored/gittensory-engine").AgentSdkQueryFn,
  *   hooks?: import("@jsonbored/gittensory-engine").AgentSdkHooks,
+ *   listChangedFiles?: (cwd: string) => Promise<string[]>,
  *   houseRulesConfig?: unknown,
  *   houseRulesOptions?: unknown,
  * }} [options]
@@ -97,5 +98,6 @@ export function constructProductionCodingAgentDriver(env, options = {}) {
     spawn: options.spawn ?? createRealCliSubprocessSpawn(),
     ...(options.query !== undefined ? { query: options.query } : {}),
     ...(hooks !== undefined ? { hooks } : {}),
+    ...(options.listChangedFiles !== undefined ? { listChangedFiles: options.listChangedFiles } : {}),
   });
 }

test/contract/coding-agent-driver-parity.test.ts

@@ -99,6 +99,9 @@ const DRIVER_HARNESSES: DriverHarness[] = [
     make: (behavior) => {
       const recorded: RecordedInvocation = { cwds: [], prompts: [] };
       const driver = createAgentSdkCodingAgentDriver({
+        // This suite exercises the shared result-shape/scoping contract, not real git enumeration; the shared
+        // WORKTREE constant is not an actual git repo, so the real default enumerator would fail closed.
+        listChangedFiles: async () => [],
         query: (input) => {
           recorded.cwds.push(input.options.cwd);
           recorded.prompts.push(input.prompt);