fix(tasks): install the gh shim at runtime for snapshot resumes

The gh PATH shim is baked into new base images, but a resume from a pre-shim
filesystem snapshot — or any window where the base image lags this backend —
would lack it, leaving gh with no token once the frozen launch-env token is
unset. Write and chmod the shim from start_agent_server (reading its single
source of truth) so it's present on every launch regardless of image age.

Author: VojtechBartos

products/tasks/backend/logic/services/agentsh.py

@@ -1,4 +1,5 @@
 import shlex
+from pathlib import Path
 from urllib.parse import urlparse
 
 from django.conf import settings
@@ -16,6 +17,19 @@
 # Sourced via BASH_ENV on every `bash -c` the agent runs, so git/gh pick up a
 # mid-session GitHub credential refresh from its dedicated credential file.
 BASH_ENV_SCRIPT = "/tmp/agentsh-bash-env.sh"
+
+# The gh PATH shim (first on PATH; sources the credential script so gh authenticates as the current
+# actor in any shell mode). Baked into new base images, but also installed at runtime so resumes
+# from pre-shim filesystem snapshots — and any window where the image lags this backend — still
+# deliver the token to gh. Read from its single source of truth (the file the Dockerfiles COPY).
+GH_GUARD_INSTALL_PATH = "/opt/posthog/bin/gh"
+_GH_GUARD_SOURCE_PATH = Path(__file__).resolve().parents[2] / "sandbox" / "images" / "gh-guard.sh"
+
+
+def read_gh_guard_script() -> bytes:
+    return _GH_GUARD_SOURCE_PATH.read_bytes()
+
+
 AGENTSH_AUDIT_DB = "/var/lib/agentsh/events.db"
 INFRASTRUCTURE_DOMAINS = [
     "*.posthog.com",

products/tasks/backend/logic/services/docker_sandbox.py

@@ -35,13 +35,15 @@
 from .agentsh import (
     BASH_ENV_SCRIPT,
     ENV_WRAPPER_SCRIPT,
+    GH_GUARD_INSTALL_PATH,
     SESSION_ID_FILE,
     build_exec_prefix,
     build_setup_script,
     generate_bash_env_script,
     generate_config_yaml,
     generate_env_wrapper,
     generate_policy_yaml,
+    read_gh_guard_script,
 )
 from .local_skills import ENV_LOCAL_SKILLS_HOST_PATH, LocalSkillsCache, bundled_skills_disabled
 from .sandbox import (
@@ -878,6 +880,15 @@ def _launch_and_check(self, command: str) -> bool:
             return False
         return self._wait_for_health_check(max_attempts=20)
 
+    def _install_gh_guard(self) -> None:
+        """Install the gh PATH shim at runtime so it's present regardless of image age.
+
+        New base images bake it in, but a resume from a pre-shim filesystem snapshot (or any window
+        where the image lags this backend) would otherwise lack it, leaving gh with no token once the
+        frozen launch-env token is unset."""
+        self.write_file(GH_GUARD_INSTALL_PATH, read_gh_guard_script())
+        self.execute(f"chmod +x {shlex.quote(GH_GUARD_INSTALL_PATH)}", timeout_seconds=30)
+
     def start_agent_server(
         self,
         repository: str | None,
@@ -926,6 +937,7 @@ def start_agent_server(
         # mid-session credential refreshes reach git/gh. Needed for both agentsh
         # and non-agentsh runs.
         self.write_file(BASH_ENV_SCRIPT, generate_bash_env_script().encode())
+        self._install_gh_guard()
 
         if allowed_domains is not None:
             self._setup_agentsh(WORKING_DIR, allowed_domains)

products/tasks/backend/logic/services/modal_sandbox.py

@@ -56,6 +56,7 @@
     AGENTSH_DAEMON_PORT,
     BASH_ENV_SCRIPT,
     ENV_WRAPPER_SCRIPT,
+    GH_GUARD_INSTALL_PATH,
     SESSION_ID_FILE,
     _hostname_from_url,
     build_exec_prefix,
@@ -64,6 +65,7 @@
     generate_config_yaml,
     generate_env_wrapper,
     generate_policy_yaml,
+    read_gh_guard_script,
 )
 from products.tasks.backend.logic.services.local_packages import (
     get_local_package_runtime_dependencies,
@@ -1106,6 +1108,11 @@ def start_agent_server(
             repo_path = f"/tmp/workspace/repos/{org}/{repo}"
 
         self.write_file(BASH_ENV_SCRIPT, generate_bash_env_script().encode())
+        # Install the gh shim at runtime too (see agentsh.GH_GUARD_INSTALL_PATH): a resume from a
+        # pre-shim filesystem snapshot — or any window where the base image lags this backend —
+        # would otherwise leave gh with no token once the frozen launch-env token is unset.
+        self.write_file(GH_GUARD_INSTALL_PATH, read_gh_guard_script())
+        self.execute(f"chmod +x {shlex.quote(GH_GUARD_INSTALL_PATH)}", timeout_seconds=30)
 
         if allowed_domains is not None:
             self._setup_agentsh(WORKING_DIR, allowed_domains)

products/tasks/backend/logic/services/tests/test_modal_sandbox.py

@@ -665,6 +665,8 @@ def test_start_agent_server_raises_on_health_check_failure(self, mock_sandbox: A
         mock_sandbox.execute = MagicMock(
             side_effect=[
                 ExecutionResult(stdout="", stderr="", exit_code=0, error=None),
+                ExecutionResult(stdout="", stderr="", exit_code=0, error=None),  # gh shim write (mv)
+                ExecutionResult(stdout="", stderr="", exit_code=0, error=None),  # gh shim chmod
                 ExecutionResult(stdout="", stderr="", exit_code=0, error=None),  # --posthogExecPermissionRegex probe
                 ExecutionResult(stdout="", stderr="", exit_code=1, error=None),
                 ExecutionResult(stdout="some log output", stderr="", exit_code=0, error=None),
@@ -720,6 +722,8 @@ def test_start_agent_server_frees_port_before_relaunch(self, mock_sandbox: Any):
         mock_sandbox.execute = MagicMock(
             side_effect=[
                 ExecutionResult(stdout="", stderr="", exit_code=0, error=None),
+                ExecutionResult(stdout="", stderr="", exit_code=0, error=None),  # gh shim write (mv)
+                ExecutionResult(stdout="", stderr="", exit_code=0, error=None),  # gh shim chmod
                 ExecutionResult(stdout="", stderr="", exit_code=0, error=None),  # --posthogExecPermissionRegex probe
                 ExecutionResult(stdout="", stderr="", exit_code=0, error=None),
                 ExecutionResult(stdout="ok:1", stderr="", exit_code=0, error=None),

products/tasks/backend/tests/test_agentsh.py

@@ -463,6 +463,8 @@ def execute(command: str, timeout_seconds: int | None = None) -> ExecutionResult
             if "--taskId" in command:
                 launched.append(command)
                 return ExecutionResult(stdout="", stderr="", exit_code=0)
+            if "chmod" in command:  # gh shim install
+                return ExecutionResult(stdout="", stderr="", exit_code=0)
             self.assertIn("grep", command)
             return ExecutionResult(stdout="", stderr="", exit_code=0 if supported else 1)