Back to skill

Security audit

Brain Memory

Security checks for vulnerabilities and agentic risk

Overview

The skill is a real persistent-memory integration, but it gives agents broad cross-session and cross-agent memory authority with weak user control and an unpinned executable install.

Install only if you intentionally want a persistent memory layer shared across agents. Treat stored and recalled memories as sensitive user data, avoid using it for secrets or regulated information, consider isolating ~/.brain per project or agent, review what is stored regularly, and prefer a pinned audited brain-memory package before deployment.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:6
Finding

Unpinned Executable npm Dependency

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:12
Finding

Cross-Agent Persistent Memory Poisoning Boundary

Content
View full analysis
` returns the budget-bounded session payload (pinned facts, relevant memories, skills index) — internalize it silently; do not dump it. - On session boundaries, append a summary entry to `~/.brain/contexts.json` (keep only the last 20) so future sessions get context-dependent recall. ``` ### Technical Analysis The Skill establishes one global memory directory shared by all AI agents and directs agents to persist conversation-derived information in that store. It also directs future sessions to silently internalize returned pinned facts, memories, and Skill-index content. The documented workflow does not specify: - Isolation between agents, users, projects, or trust domains. - Validation that stored memory content contains facts rather than instructions. - Prompt-injection filtering when recalling Markdown memory bodies. - Strong provenance checks before future sessions internalize stored content. - User review of ordinary persistent writes when memorization has been requested. - A rule pr ...[truncated 2795 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill instructs the agent to store durable user preferences, decisions, learnings, and life-domain details immediately and to classify them itself, without confirmation when the user asked to memorize. This materially increases the risk of collecting and persisting sensitive personal data, including information about family, social relationships, health routines, and professional matters, in a long-lived store.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill establishes a single global ~/.brain directory shared across all AI agents and instructs later recall of stored memories. In context, this creates a cross-session and potentially cross-agent data exposure channel for personal information, increasing the likelihood of unintended disclosure, over-collection, and misuse.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
75% confidence
Finding

Although the matched snippet is somewhat noisy, the surrounding skill behavior clearly describes persistent memory storage and reuse across sessions, which constitutes session persistence risk. In this skill's context, that persistence is a core feature, but it still creates privacy and data leakage concerns because recalled information may surface in later interactions unexpectedly.

Content

Scanner excerpt · SKILL.md (reported line 15)May include surrounding context.

md
Markdown files with YAML frontmatter (type, cognitive type, strength, decay,
salience, confidence, tags, associations). Recalled memories get stronger;
ignored ones fade. Use the `brain` CLI for every operation — never compute
scores or write memory files by hand.

## Recall (when asked to "remember", or when past context would help)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly instructs the agent to store user information immediately and report after, without a user-facing privacy warning or consent checkpoint. Because the data includes preferences, decisions, relationships, and other life-domain details in a persistent shared store, this creates a meaningful retention and privacy risk across future sessions and agents.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill tells the agent to silently internalize session payloads and append cross-session summaries, enabling background accumulation and reuse of user context. The silent nature of this ingestion makes the retention less transparent and raises the risk of hidden profiling and unintended downstream disclosure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The session instructions describe appending summaries to a persistent contexts file without telling the user that conversation-derived data will continue to be retained. Silent session-summary persistence increases the chance users disclose information believing it is ephemeral when it is actually stored for later reuse.

Content

No source excerpt is available for this finding.