Back to skill

Security audit

bluecolumn-memory

Security checks for vulnerabilities and agentic risk

Overview

This is a real remote-memory skill, but it should be reviewed carefully because it encourages persistent storage of broad raw user content on an external service without enough scoping or privacy controls.

Install only if you intend to use BlueColumn as a third-party persistent memory service. Do not store secrets, credentials, regulated data, confidential transcripts, or full documents unless you have authorization and understand BlueColumn's retention and deletion controls. Prefer pinned package versions or reviewed source when using the npx install or MCP examples.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (13)

Rp1

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The README instructs users to execute npx skills add ... without pinning a specific package version. This creates a supply-chain risk because future package updates or a compromised published version could change what gets executed at install time, and users will fetch the latest release implicitly.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README promotes storing user facts, decisions, preferences, and audio-derived content in a remote memory service but does not clearly warn that potentially sensitive user data will leave the local agent environment. In an agent-skill context, this omission increases the chance that operators will send confidential or regulated data off-host without informed consent or proper data-handling review.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The README recommends running npx bluecolumn-mcp --api-key=... without a pinned version. Unpinned npx execution can pull and run newly published code automatically, which is dangerous if the package is compromised, typosquatted, or changed in a future release.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
87% confidence
Finding

The example demonstrates sending user profile information to https://api.bluecolumn.ai/remember, which is an external transmission of potentially sensitive data. Although this is the intended function of the skill, it is still security-relevant because operators may copy the example without realizing they are exporting user memory and preferences to a third-party service.

Content

Scanner excerpt · README.md (reported line 47)May include surrounding context.

bash
# Store
curl -X POST https://api.bluecolumn.ai/remember \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer bc_live_YOUR_KEY" \
  -d '{"text": "User is building a CRM agent; prefers TypeScript", "title": "User profile"}'

External Transmission

Medium
Category
Data Exfiltration
Confidence
87% confidence
Finding

The example demonstrates sending user profile information to https://api.bluecolumn.ai/remember, which is an external transmission of potentially sensitive data. Although this is the intended function of the skill, it is still security-relevant because operators may copy the example without realizing they are exporting user memory and preferences to a third-party service.

Content

Scanner excerpt · README.md (reported line 47)May include surrounding context.

bash
# Store
curl -X POST https://api.bluecolumn.ai/remember \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer bc_live_YOUR_KEY" \
  -d '{"text": "User is building a CRM agent; prefers TypeScript", "title": "User profile"}'

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 53)May include surrounding context.

md
-d '{"text": "User is building a CRM agent; prefers TypeScript", "title": "User profile"}'

# Recall
curl -X POST https://api.bluecolumn.ai/recall \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer bc_live_YOUR_KEY" \
  -d '{"q": "What is the user building?"}'

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

MCP server references in the skill manifest without version pinning are a rug-pull risk.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill prominently encourages persistent, searchable memory storage but does not clearly warn that user content will be sent to a third-party service and retained there. In an agent context, this can cause inadvertent exfiltration of sensitive prompts, transcripts, secrets, or personal data because operators may assume memory remains local unless explicitly told otherwise.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill encourages storing broad categories of information including facts, decisions, preferences, promises, action items, and transcripts in persistent memory. In practice, that creates a natural-language data retention channel where secrets, regulated data, or private conversation content may be uploaded and later recalled outside the user's expectations.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The /remember example explicitly tells agents to store 'Full raw content to store (transcript, doc, note...)', which strongly promotes bulk persistence of potentially sensitive material. Because agents often process emails, chats, documents, and meeting notes, this pattern can easily lead to unauthorized retention and later disclosure of confidential data via recall queries.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The usage example instructs storing a full transcript in persistent memory, normalizing a risky behavior that can capture sensitive conversation content wholesale. In an agent setting, transcripts may contain access tokens, legal discussions, customer data, or internal strategy, making later retrieval a meaningful confidentiality risk.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

The skill recommends npx bluecolumn-mcp without a version pin, which exposes users to supply-chain risk if the published package is updated maliciously or unexpectedly. Because this launches an MCP server that may handle agent data and tool access, executing an unpinned package is more sensitive than a passive documentation link.

Content

No source excerpt is available for this finding.