Back to skill

Security audit

Memkeeper for OpenClaw

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent for setting up local Memkeeper memory, but it tells the agent to download and run an unpinned remote installer script.

Review the installer source before use, prefer a pinned release or verified checksum/signature, and install only if you are comfortable registering a persistent local MCP memory server that stores user-level memory outside the current project by default.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:21
Finding

Unverified Remote Installer Download and Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 21–28
Vulnerability Type: Remote payload retrieval and execution
Risk Level: High

Vulnerable Code

bash
If `MEMKEEPER_BIN` is absent, download the public installer to a temporary file,
inspect it, and run it. Do not paste secrets into a command or config file.

```bash
curl -fsSL https://raw.githubusercontent.com/teflon07/memkeeper/main/install.sh \
  -o /tmp/memkeeper-install.sh
bash /tmp/memkeeper-install.sh
text

### Technical Analysis

The Skill instructs the agent to download a shell script from a personal GitHub repository's mutable `main` branch and execute it with `bash`. It does not pin the download to an immutable commit or release, nor does it verify a checksum or cryptographic signature.

Although the prose says to inspect the installer, the provided command sequence contains no inspection, integrity check, or explicit approval gate. Consequently, the payload that executes can differ from the content reviewed during the Skill audit. Using the predictable `/tmp/memkeeper-install.sh` path also provides weaker temporary-file handling than a securely created, owner-restricted temporary directory.

Installing Memkeeper is relevant to the declared functionality, but granting a mutable remote script arbitrary command execution exceeds the minimum authority required. The later commands trust the resulting executable to download models, operate on the user's memory store, and act as an OpenClaw MCP server.

### Attack Path

1. An attacker compromises or otherwise gains control of the upstream GitHub account, repository, or mutable `main` branch.
2. The attacker modifies `install.sh` to include malicious shell commands while preserving expected installation behavior.
3. A user or agent follows the Skill when `MEMKEEPER_BIN` is absent.
4. `curl` downloads the altered installer without validating a pinned digest or signature.
5. `bash /t
...[truncated 1178 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace the mutable main-branch URL with an official, immutable, versioned release artifact or commit-pinned source.
  2. Publish and verify a SHA-256 or stronger digest before executing or installing anything. Prefer cryptographic signature verification using a separately distributed, trusted public key.
  3. Avoid executing remote shell installers where a signed package or directly verifiable binary is available.
  4. If an installer is unavoidable, vendor a reviewed copy in the Skill package or pin its exact content, inspect it before use, and require explicit operator approval before execution.
  5. Fail closed if the version, checksum, signature, origin, or expected file type cannot be verified.
  6. Create temporary files with mktemp inside an owner-restricted directory, apply restrictive permissions, and remove them after installation.
  7. Run installation with the least-privileged user and reject unexpected requests for sudo, system-level service installation, or access outside documented directories.
  8. Verify the installed binary's digest before registering it as an MCP server, and constrain its executable and filesystem permissions where practical.
  9. Pin and validate model artifacts downloaded by pull-models when the upstream supports integrity metadata.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep