Security audit
Security checks for vulnerabilities and agentic risk
The skill openly patches local OpenClaw installation files for a stated prompt-cache behavior change and provides dry-run, backup, validation, and rollback controls.
Install only if you are comfortable editing your local OpenClaw installation. Run the dry-run first, confirm the target root and bundle paths, consider testing with --root on a copied fixture, and keep the rollback script available before restarting the gateway.
Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.
#!/usr/bin/env python3
"""Shared helpers for the responses third-party prompt-cache patch skill."""
from __future__ import annotations
The skill explicitly instructs users to run local patch and rollback scripts that write into an installed OpenClaw dist/ bundle and invoke shell commands, but it declares no permissions or allowed-tools scope. That mismatch is dangerous because a consumer or agent may execute file-write and shell-capable actions without an explicit trust boundary, increasing the risk of unintended modification of a live installation.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def _resolve_root_from_systemctl() -> tuple[str, Path] | None:
try:
result = subprocess.run(
["systemctl", "--user", "show", SYSTEMD_UNIT_NAME, "--property=ExecStart", "--value"],
capture_output=True,
text=True,
This helper writes directly to bundle files and creates backup copies via write_text and create_backup, which are safety-relevant filesystem modifications. Although errors are surfaced, the file contains no confirmation prompt before modification and no user-facing disclosure in comments or docstrings describing that installed files will be changed.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def run_node_check(path: Path) -> None:
try:
result = subprocess.run(
["node", "--check", str(path)],
capture_output=True,
text=True,
No suspicious patterns detected.