From 41d7dc38781c68f26fe124d5c20c96aa1319bdb8 Mon Sep 17 00:00:00 2001
From: adarsh tiwari <adarshtiwari908@gmail.com>
Date: Wed, 3 Jun 2026 16:55:56 +0530
Subject: [PATCH] feat: add Stop hook example policy (closes #40)

New file: examples/policies-stop.js

Demonstrates intercepting the Stop event to:

- Block the stop if there are uncommitted git changes

- Allow the stop and log a task-completion summary otherwise

Both code paths have inline comments explaining *why* each decision is made.
---
 examples/policies-stop.js | 65 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 65 insertions(+)
 create mode 100644 examples/policies-stop.js

diff --git a/examples/policies-stop.js b/examples/policies-stop.js
new file mode 100644
index 00000000..ef74fa92
--- /dev/null
+++ b/examples/policies-stop.js
@@ -0,0 +1,65 @@
+/**
+ * policies-stop.js — Stop event hook example
+ *
+ * Demonstrates intercepting the `Stop` event that fires when the AI agent
+ * finishes a task and is about to exit. The hook enforces a pre-exit gate:
+ *   - BLOCK the stop if there are uncommitted git changes
+ *   - ALLOW the stop (with a logged summary) when the working tree is clean
+ *
+ * Install:
+ *   failproofai --install-hooks custom ./examples/policies-stop.js
+ *
+ * Test:
+ *   1. Make a dirty working tree (e.g. `echo x >> scratch.txt`) and let Claude
+ *      finish a task. The hook will deny the stop with a human-readable reason.
+ *   2. Commit or stash everything, then let Claude finish — the hook will allow
+ *      the stop and append a line to `task-summary.log`.
+ */
+import { customPolicies, allow, deny } from "failproofai";
+import { execSync } from "child_process";
+import fs from "fs";
+
+customPolicies.add({
+  name: "enforce-stop-cleanup-and-checks",
+  description:
+    "Gate the agent's exit: block if uncommitted changes exist, allow and log a summary otherwise.",
+  match: { events: ["Stop"] },
+  fn: async (ctx) => {
+    // ── "Block the stop" path ─────────────────────────────────────────
+    // WHY: An AI agent that exits with uncommitted changes leaves the
+    //      developer in an ambiguous state — did the task succeed?  Were
+    //      the edits intentional?  Blocking the stop forces the agent to
+    //      commit or stash first, producing a clean audit trail.
+    try {
+      const status = execSync("git status --porcelain", { encoding: "utf8" });
+      if (status.trim().length > 0) {
+        return deny(
+          "Uncommitted changes detected. Please commit or stash before stopping."
+        );
+      }
+    } catch {
+      // If git is unavailable or the cwd isn't a repo, skip the check
+      // rather than crashing the hook — graceful degradation is important
+      // so the hook doesn't block every non-git project.
+    }
+
+    // ── "Let it stop" path ────────────────────────────────────────────
+    // WHY: When the working tree is clean we know the task's output has
+    //      been captured in a commit.  We take a side-effect action
+    //      (appending to a log file) so the team has an out-of-band
+    //      record that the session completed, then allow the exit.
+    try {
+      const timestamp = new Date().toISOString();
+      const sessionId = ctx.session?.sessionId ?? "unknown";
+      fs.appendFileSync(
+        "task-summary.log",
+        `[${timestamp}] session=${sessionId} — task completed, stop allowed\n`
+      );
+    } catch {
+      // Logging is best-effort; never block the agent just because the
+      // log file couldn't be written.
+    }
+
+    return allow();
+  },
+});