Back to skill

Security audit

Memory System Complete

Security checks for vulnerabilities and agentic risk

Overview

This is mainly a local memory database skill, but it needs Review because its optional embedding path can send stored memories to any configured Ollama endpoint while the docs promise data stays local.

Install only if you are comfortable with a local persistent memory database that can create, update, export, and delete records. Keep Ollama pointed at localhost unless you explicitly trust the endpoint, because semantic search may send memory contents to it. Back up the database before cleanup, force reinitialization, import/export, or upgrade operations, and use a virtual environment with pinned dependencies where possible.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/memory_system.py:47
Finding

Configurable Ollama Endpoint Can Disclose Stored Memory Content

Content
View full analysis
List[float]: """Generate embedding""" try: payload = {"model": self.model, "prompt": text} response = requests.post(self.api_url, json=payload, timeout=30) result = response.json() ``` ### Technical Analysis The `ollama_url` configuration value is accepted without validating the URL scheme, hostname, resolved address, port, or redirect behavior. Although the default value points to localhost, callers can configure any HTTP or HTTPS endpoint. When semantic search is enabled, the implementation retrieves up to 1,000 memories and sends each complete title ...[truncated 2435 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:357
Finding

Installation Guidance Uses Unpinned Third-Party Dependencies

Content
View full analysis
=1.20.0 # For graph analysis pip install networkx>=2.0 # For vector search pip install lancedb>=0.3.0 pip install sentence-transformers>=2.0.0 ``` ```bash # SKILL.md:583 pip install networkx ``` ```bash # README.md:303 pip install numpy>=1.20.0 ``` ### Technical Analysis The installation documentation instructs users to install mutable dependency ranges or completely unpinned packages. The project does not provide a lock file, exact reviewed versions, package hashes, or an enforced trusted package index. As a result, installation behavior can change after the Skill has been audited. Pip may resolve a newer, incompatible, or compromised release. Python packages can execute code during build and installation, so a compromised dependency can run with the privileges of the user performing the installation. No evidence was found that the currently named packages are malicious. The confirmed issue is the absence of reproducible, integrity-verified dependency controls. ### Attack Path 1. A user follows the documented installation instructions. 2. Pip contacts the configured package index, which may be the public index, a corporate mirror, or a user-configured untrusted source. 3. Because the dependency is unpinned or specified only with a lower bound, pip selects a package version that was not necessarily reviewed with this Skill. 4. An upstream account compromise, malicious mirror, dependency takeover, or future compromised release supplies hostile package content. 5. Pip downloads and builds or installs the selected package. 6. Package build or installation code executes with the installing user’s privileges. 7. The malicious package can access da ...[truncated 822 chars]
Remediation
View remediation
Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (54)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

A verification utility that inserts and deletes records is more than passive validation and should be disclosed as database modification behavior. In a memory-management context, unexpected writes or deletions are more dangerous because users may assume their local data store remains untouched except for deliberate memory operations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The privacy section states that all memory data stays local and is not uploaded, yet the same document advertises network search and other network-capable integrations. This contradiction is dangerous because users may entrust sensitive memory data under false assumptions and the skill may later transmit memory-derived content, queries, or context externally.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This is a markdown file, so SQP-2 applies to omissions in the skill description. The file advertises automatic cleanup, CRUD, and import/export capabilities, and later shows cleanup deleting memories, but it does not warn users that stored memory data may be modified or removed.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The example explicitly cleans up and reports deleted memories, indicating potentially irreversible deletion. In markdown, safety-relevant behaviors that affect user data should be accompanied by warnings, but none are provided here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file presents all user-facing instructions and installation guidance exclusively in Chinese. Under the policy for natural-language violations, forcing a specific language without user opt-in or a documented locale justification is reportable.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document explicitly advertises Theory-of-Mind, belief updates, intent inference, and emotion detection on user content, but provides no notice, consent mechanism, privacy limitation, or warning about profiling-sensitive analysis. Even though this is a release note rather than executable code, it describes functionality that can infer internal states and preferences from user text, which creates privacy, trust, and potential policy/compliance risks if deployed without transparency and guardrails.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The document explicitly describes persistent storage of sensitive user-derived artifacts such as memories, retrieval history, review queue entries, deep research data, ingestion cache, and evolution logs, but gives no privacy notice, retention guidance, consent expectations, or warning about persistence. In a memory-system skill, this omission is security-relevant because users may provide personal or confidential content, and the skill context increases the likelihood that long-lived behavioral and knowledge data will accumulate silently.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README instructs users to initialize and force reinitialize a local database, but it does not clearly warn that these operations may overwrite, reset, or otherwise modify existing local memory data. In a skill explicitly designed to store persistent user memories, omission of a data-loss warning increases the risk that users will unintentionally destroy sensitive or valuable local data.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The manifest defines post-install commands that create directories and run Python scripts, and the documentation also advertises optional network-capable components such as Ollama and deep research, yet no explicit permissions or allowed-tools scope is declared. This is dangerous because installers and agents may execute filesystem or network-capable actions without clear user consent boundaries, increasing the chance of unexpected writes or outbound access.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Omitting documented deep research and multi-query network search from the manifest is security-relevant because it conceals network-capable behavior from the primary metadata users rely on during installation. Hidden search/network activity can expose user prompts or memory-derived queries outside the local environment.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

Omitting documented deep research and multi-query network search from the manifest is security-relevant because it conceals network-capable behavior from the primary metadata users rely on during installation. Hidden search/network activity can expose user prompts or memory-derived queries outside the local environment.

Content

No source excerpt is available for this finding.