test: stop the suite writing to the real ~/.pi/agent audit log

tests/audit.test.ts

@@ -8,6 +8,7 @@
  * clear, tail (with corruption defense), count, + the concurrency guarantee
  * (parallel appends don't interleave or corrupt — T68).
  */
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { mkdtempSync, rmSync, existsSync, readFileSync, writeFileSync, statSync, chmodSync } from "node:fs";

tests/batch.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { mapWithConcurrency } from "../lib/batch.ts";

tests/cache.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { mkdtempSync, rmSync, utimesSync, writeFileSync, existsSync } from "node:fs";

tests/capability.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import type { Api, Model } from "@earendil-works/pi-ai";

tests/config.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { mkdtempSync, readFileSync, readdirSync, rmSync, writeFileSync } from "node:fs";

tests/defaults.test.ts

@@ -8,6 +8,7 @@
  * (sorted by `(provider, id)` so the registry's iteration order doesn't
  * matter).
  */
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import type { Api, Model } from "@earendil-works/pi-ai";

tests/delegate.test.ts

@@ -1,10 +1,12 @@
+import { isInsideRealPiDir, REAL_PI_DIR } from "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
-import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { mkdtempSync, readdirSync, rmSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
-import { join } from "node:path";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
 import type { Api, Model } from "@earendil-works/pi-ai";
-import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
+import { getAgentDir, type ExtensionContext } from "@earendil-works/pi-coding-agent";
 import {
   callVisionModel,
   delegateToVisionModel,
@@ -620,8 +622,8 @@ test("delegateToVisionModel: abort → code 'aborted', 0 calls, no fallback", as
 });
 // ── v0.5.0 (SPEC-5) tests: audit log + local-only mode ───────────────────
 
-import { readFileSync, existsSync, mkdirSync } from "node:fs";
-import { countAuditLog, tailAuditLog, clearAuditLog } from "../lib/audit.ts";
+import { readFileSync, existsSync } from "node:fs";
+import { countAuditLog, clearAuditLog } from "../lib/audit.ts";
 
 /** Helper: a temp dir + an image file + a configured ctx + an audit log path. */
 function setupAuditTest() {
@@ -881,3 +883,41 @@ test("T47 regression: single-image success with audit on → v0.4.0 behavior pre
 });
 
 
+
+// ── Guard: the agent-dir redirect from tests/setup.ts must stay in place ───
+// Tests that leave auditLogPath unset resolve to getAgentDir(). Without the
+// redirect, every one of them appends a line to the developer's real
+// ~/.pi/agent/vision-audit.log — silently mixing fixture entries (fake
+// providers, images that never existed) into the log that answers "where did
+// my image bytes actually go?".
+test("agent dir is redirected away from the real ~/.pi/agent", () => {
+  const agentDir = getAgentDir();
+  assert.ok(
+    process.env.PI_CODING_AGENT_DIR,
+    `agent dir must follow the redirect — is ./setup.ts still the first import?`,
+  );
+  assert.ok(
+    !isInsideRealPiDir(agentDir),
+    `agent dir must not be inside ${REAL_PI_DIR}, got ${agentDir}`,
+  );
+});
+
+// The redirect only holds for a file that installs it, and `tsx --test <file>`
+// is the normal single-file loop — so a new test file that forgets the import
+// silently writes to the real log. Cheaper to catch here than in the log.
+test("every test file installs the agent-dir redirect first", () => {
+  const testsDir = dirname(fileURLToPath(import.meta.url));
+  const files = readdirSync(testsDir).filter((f) => f.endsWith(".test.ts"));
+  assert.ok(files.length > 1, "expected to find the test files");
+  for (const file of files) {
+    const source = readFileSync(join(testsDir, file), "utf8");
+    const firstImport = source
+      .split("\n")
+      .find((line) => /^import[\s{]/.test(line));
+    assert.match(
+      firstImport ?? "",
+      /^import (?:.+ from )?"\.\/setup\.ts";$/,
+      `${file} must import "./setup.ts" before anything else — otherwise it writes to the real ~/.pi/agent`,
+    );
+  }
+});

tests/image.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { mkdtempSync, rmSync, writeFileSync, mkdirSync } from "node:fs";

tests/integration.test.ts

@@ -12,6 +12,7 @@
  * reason about the attached image / choose to call describe_image) still
  * needs a manual fresh-session check — see the session handoff.
  */
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
@@ -25,18 +26,17 @@ import type {
   ExtensionContext,
   ToolDefinition,
 } from "@earendil-works/pi-coding-agent";
-
-// Redirect getAgentDir() to a temp dir so /vision commands never touch the
-// real ~/.pi/agent/vision.json during this test run.
-const TMP_AGENT = mkdtempSync(join(tmpdir(), "vision-eval-agent-"));
-process.env.PI_CODING_AGENT_DIR = TMP_AGENT;
-
 import visionFactory from "../extensions/vision.ts";
 import pasteFactory from "../extensions/paste.ts";
 import { loadConfig, configFilePath } from "../lib/config.ts";
 import { getAgentDir } from "@earendil-works/pi-coding-agent";
 import { countAuditLog, tailAuditLog } from "../lib/audit.ts";
 
+// The redirect away from the real ~/.pi/agent is installed by ./setup.ts, which
+// also owns the temp dir's lifetime; this is just the dir the /vision commands
+// under test read and write.
+const TMP_AGENT = getAgentDir();
+
 const PNG_1x1_B64 =
   "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAC0lEQVR42mNk+M8AAAMBEg1+mP0AAAAASUVORK5CYII=";
 const PNG_BYTES = Buffer.from(PNG_1x1_B64, "base64");
@@ -1982,8 +1982,3 @@ test("T70: v0.5.0 regression gate — config fields + subcommands + auto-detect
   // Full suite green = T70 passed (if this test runs, the suite compiled + loaded).
   assert.ok(true, "v0.5.0 surface wired + regression gate passed");
 });
-
-// Cleanup the temp agent dir after all tests.
-test("cleanup", () => {
-  rmSync(TMP_AGENT, { recursive: true, force: true });
-});

tests/marker.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import {

tests/normalize.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { normalizeImagePaths } from "../extensions/vision.ts";

tests/paste.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { findImagePathTokens } from "../extensions/paste.ts";

tests/preview.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import {

tests/resilience.test.ts

@@ -1,3 +1,4 @@
+import "./setup.ts";
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { AbortError, classifyError, sleep, withRetry, type ErrorClass } from "../lib/resilience.ts";

tests/setup.ts

@@ -0,0 +1,71 @@
+// Agent-dir redirect for the test suite — imported as the first statement of
+// every tests/*.test.ts file, so it takes effect however the suite is started:
+// `pnpm test`, a bare `tsx --test`, or a single file. (A meta-test in
+// delegate.test.ts enforces that every test file still does so.)
+//
+// Why: config, cache and the audit log all resolve through getAgentDir(), which
+// reads PI_CODING_AGENT_DIR at call time and otherwise falls back to the real
+// ~/.pi/agent. Any test that delegates with the default config (auditLog is on)
+// would append fixture rows — fake providers, images that never existed — to the
+// developer's production vision-audit.log.
+import { mkdirSync, mkdtempSync, rmSync } from "node:fs";
+import { homedir, tmpdir } from "node:os";
+import { join, resolve, sep } from "node:path";
+
+/** The real pi config root — nothing the suite writes may land inside it. */
+export const REAL_PI_DIR = join(homedir(), ".pi");
+
+/** Mirror of the tilde handling getAgentDir() applies to the env var, so the
+ *  value we hand on is the one getAgentDir() will return. */
+function expand(dir: string): string {
+  if (dir === "~") return homedir();
+  if (dir.startsWith("~/")) return join(homedir(), dir.slice(2));
+  return resolve(dir);
+}
+
+/** True if `dir` is ~/.pi itself or anything below it. Separator-aware: a
+ *  sibling such as ~/.pi-sandbox is a legitimate test dir, not the real one. */
+export function isInsideRealPiDir(dir: string): boolean {
+  const abs = expand(dir);
+  return abs === REAL_PI_DIR || abs.startsWith(REAL_PI_DIR + sep);
+}
+
+// Respect a dir supplied from outside (CI, a wrapper script) — but never one
+// that points back into ~/.pi, which is exactly the pollution this file exists
+// to prevent. Only the dir we created ourselves is ours to delete.
+let ownDir: string | undefined;
+const inherited = process.env.PI_CODING_AGENT_DIR;
+
+if (inherited && !isInsideRealPiDir(inherited)) {
+  process.env.PI_CODING_AGENT_DIR = expand(inherited);
+  mkdirSync(process.env.PI_CODING_AGENT_DIR, { recursive: true });
+} else {
+  if (inherited) {
+    console.warn(
+      `[tests/setup] ignoring PI_CODING_AGENT_DIR=${inherited} — it resolves inside ${REAL_PI_DIR}; using a temp dir instead`,
+    );
+  }
+  ownDir = mkdtempSync(join(tmpdir(), "vision-test-agent-"));
+  process.env.PI_CODING_AGENT_DIR = ownDir;
+}
+
+function cleanup(): void {
+  if (ownDir) rmSync(ownDir, { recursive: true, force: true });
+}
+
+process.on("exit", cleanup);
+
+// `exit` never fires when we are terminated by a signal — a Ctrl-C on the suite
+// reaches every per-file child process and would otherwise leak one temp dir
+// each. rmSync({ force: true }) is idempotent, so the extra `exit` pass that
+// process.exit() triggers is harmless.
+for (const [signal, code] of [
+  ["SIGINT", 130],
+  ["SIGTERM", 143],
+  ["SIGHUP", 129],
+] as const) {
+  process.on(signal, () => {
+    cleanup();
+    process.exit(code);
+  });
+}