ClawHub
Back to skill

Security audit

Session Memory Flush

Security checks for vulnerabilities and agentic risk

Overview

This skill does what it says, but it needs review because it persistently scans private OpenClaw sessions and may send transcript contents to an LLM provider.

Install only if you want a recurring background process to scan all OpenClaw-visible sessions, summarize transcripts, write memory files, and potentially send transcript text to your configured LLM provider. Use a trusted endpoint and least-privilege key, run the dry-run first, inspect the generated systemd timer or crontab, and uninstall it when you no longer want automatic scanning.

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Tainted flow: 'cmd' from os.environ.get (line 273, credential/environment) → subprocess.run (code execution)

Medium
Category
Data Flow
Content
raise RuntimeError("openclaw command not found; set OPENCLAW_BIN to the absolute path")

    cmd = [openclaw_bin, "sessions", "--all-agents", "--json"]
    proc = subprocess.run(
        cmd,
        text=True,
        stdout=subprocess.PIPE,
Confidence
86% confidence
Finding
The executable path comes from OPENCLAW_BIN and is executed without trust validation, so any actor able to influence the environment can cause this skill to run an arbitrary program. In an agent/runtime context, environment variables are often easier to tamper with than source code, making this a realistic command-execution primitive.

Tainted flow: 'req' from os.environ.get (line 614, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Content
method="POST",
    )
    try:
        with urllib.request.urlopen(req, timeout=60) as resp:
            data = json.loads(resp.read().decode("utf-8"))
    except (urllib.error.URLError, TimeoutError, json.JSONDecodeError):
        return None
Confidence
98% confidence
Finding
This code sends transcript contents plus an API key to a base URL that can be fully controlled via environment variables, enabling exfiltration of sensitive conversation data to an attacker-controlled endpoint. Because the transcript may contain secrets, internal instructions, or user data, the skill context makes this especially dangerous.

Tainted flow: 'req' from os.environ.get (line 614, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Content
method="POST",
    )
    try:
        with urllib.request.urlopen(req, timeout=90) as resp:
            body = resp.read().decode("utf-8", errors="replace")
    except (urllib.error.URLError, TimeoutError):
        return None
Confidence
98% confidence
Finding
The same untrusted endpoint pattern appears in the responses API path: environment-controlled base_url and credentials are used to POST transcript data externally. This creates a direct data-exfiltration channel and may also leak the bearer token to a malicious service.

Lp3

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding
The skill explicitly describes capabilities to read session transcripts, access environment/config, write memory files, invoke shell commands, and send transcript content to an LLM provider, yet it declares no permissions. That mismatch creates a real security risk because operators and policy engines cannot accurately assess or constrain the skill’s access, and sensitive transcript data may be exfiltrated or modified under a misleadingly low-trust manifest.

Vague Triggers

Medium
Confidence
82% confidence
Finding
The description and usage language make the skill applicable to many session types and lifecycle moments without strong trigger constraints or operator confirmation. Because the skill reads visible sessions and their transcripts, broad invocation increases the chance of unintended collection, summarization, and persistence of sensitive context from sessions that the user did not intend to flush.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The skill transmits session transcripts to an external LLM service without any user-facing disclosure or consent flow in this file. Given the stated purpose of summarizing multiple sessions before reset, the data can include high-value contextual memory, making silent export materially risky.

Static analysis

No suspicious patterns detected.