Back to skill

Security audit

SolCore Memory System

Security checks for vulnerabilities and agentic risk

Overview

This is a real persistent-memory plugin, but it stores and retrieves raw conversation content with weak privacy, scoping, and trust-boundary controls.

Install only if you intentionally want long-term memory of conversations and can control the SolCore webhook and PostgreSQL database. Avoid storing secrets, credentials, regulated data, or private business content until the plugin documents consent, redaction, retention, deletion, endpoint validation, and trust handling for retrieved memories.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T02 · Agent Memory Poisoning

Error
Location
index.ts:64
Finding

Persistent Memory Poisoning Through Untrusted Stored and Retrieved Content

Content
View full analysis
{ try { const result = await callSolCore('/context', { query: args.query, limit: args.limit || 5, include_patterns: args.include_patterns !== false }, webhookUrl); return { content: [ { type: "text", text: `Context retrieved for "${args.query}":\n\n${result.context || "No relevant memories found."}` } ] }; ``` ```ts execute: async (args: { input: string; output: string; session_id?: string; user_id?: string }) => { try { await callSolCore('/store', { input: args.input, output: args.output, session_id: args.session_id, user_id: args.user_id || "michael" }, webhookUrl); ``` ### Technical Analysis The storage tool accepts arbitrary user input and assistant output and forwards both values to the persistent memory service. The retrieval tool subsequently inserts `result.context` directly into tool output without applying trust labeling, content validation, sanitization, provenance checks, or separation between instructions and reference data. Consequently, instruction-like text supplied by an attacker can be persisted and later returned as agent context. If the consuming agent treats retrieved memories as authoritative instructions rather than untrusted historical data, the content can alter future responses across sessions. The vulnerable package does not itself implement the external memory server, so server-side access controls and filtering cannot be assessed. Nevertheless, the client-side data flow provides no safeguards against persistent indirect prompt injection. ### Attack Path 1. An attacker submits user content containing ...[truncated 1279 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
index.ts:9
Finding

Unrestricted Webhook Destination Can Expose Conversation Data and Enable Server-Side Requests

Content
View full analysis
; const webhookUrl = cfg.webhookUrl; return { webhookUrl: typeof webhookUrl === "string" ? webhookUrl : "http://localhost:5003", }; } async function callSolCore(endpoint: string, payload: any, webhookUrl: string): Promise { const response = await fetch(`${webhookUrl}${endpoint}`, { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify(payload) }); if (!response.ok) { throw new Error(`SolCore error: ${response.status}`); } return await response.json(); } ``` ```ts register(api: any) { const webhookUrl = parsePluginConfig(api.pluginConfig).webhookUrl || "http://localhost:5003"; ``` The storage call sends complete interaction content to this destination: ```ts await callSolCore('/store', { input: args.input, output: args.output, session_id: args.session_id, user_id: args.user_id || "michael" }, webhookUrl); ``` ### Technical Analysis `webhookUrl` is accepted as any string and concatenated with fixed endpoint paths before being passed to `fetch`. The code performs no scheme validation, hostname allowlisting, destination resolution checks, transport-security enforcement, or authentication. A configuration value can therefore redirect storage and context requests to an arbitrary HTTP or HTTPS destination reachable from the OpenClaw host. Because the payload can contain complete user input, assistant output, session identifiers, and user identifiers, a malicious or incorrectly configured endpoint can collect sensitive conversation data. The default endpoint also uses plaintext ...[truncated 2057 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The README instructs agents to store full user input and full model output after responding, which can capture secrets, credentials, financial details, health information, or other sensitive content without any warning or filtering requirement. Because this skill is specifically designed for persistent memory and pattern detection, the omission is more dangerous than in a transient logging context: it encourages systematic retention of sensitive conversational data.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This tool persists both user input and assistant output by POSTing them to a webhook, again without any clear warning, consent, or controls around retention, scope, or sensitivity of stored data. Storing full conversations creates a stronger privacy and security risk than transient retrieval because it can capture secrets, personal data, and sensitive outputs for later misuse or compromise.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README explicitly describes persistent storage of user behavior, emotional signals, decisions, and interaction content, but provides no privacy notice, consent model, retention limits, or guidance for handling sensitive data. In a memory plugin for an AI agent, this creates a real risk of over-collection and long-term retention of highly sensitive personal and behavioral data beyond user expectations.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
72% confidence
Finding

The skill is intentionally built around persistent session memory backed by PostgreSQL, so session persistence is not accidental; however, it still expands the attack and privacy surface by retaining conversational state across interactions. In this context, persistence becomes riskier because the stored content includes behavioral, emotional, and decision-related data, increasing the consequences of unauthorized access, misuse, or excessive retention.

Content

Scanner excerpt · README.md (reported line 94)May include surrounding context.

Database Setup

bash
# Create database
createdb solcore

# The webhook will auto-migrate schema on first run

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly claims it 'automatically' stores every interaction, which can include sensitive prompts, personal data, credentials, or proprietary information, yet the documentation provides no warning, consent language, retention details, or guidance on limiting collection. In a persistent memory system, this omission is security-relevant because users may unknowingly enable broad long-term logging to PostgreSQL, increasing exposure in the event of misuse, overcollection, or database compromise.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The tool sends the user's query to an external/local webhook service to retrieve context, but the code provides no explicit notice, consent flow, or data handling restriction before transmitting potentially sensitive user content. Because the webhook URL is configurable, this can route prompts to arbitrary endpoints, increasing privacy and data exfiltration risk beyond an expected in-process memory lookup.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The plugin explicitly defines a persistent memory tool that stores both raw user input and assistant output, but the manifest provides no indication of notice, consent, retention limits, or data minimization. Because this is a memory system by design, it is likely to accumulate sensitive prompts, personal data, credentials, or proprietary information, creating privacy and compliance risk if users are unaware of persistence.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

This manifest description says the skill is for a persistent memory system for behavioral tracking and pattern detection, but it does not clarify when the skill should be invoked versus when it should not. In a manifest file, this kind of broad capability description can contribute to ambiguous triggering because there are no explicit constraints, trigger phrases, or exclusion conditions.

Content

No source excerpt is available for this finding.