Back to skill

Security audit

supermemory

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marked this memory-management skill suspicious because scanners report shell-command injection, unsafe handling of stored secrets and memories, and other capabilities or guidance beyond its stated purpose.

Consider avoiding installation unless you need this skill and can review or mitigate the reported risks. In particular, consider whether you can prevent untrusted inputs from reaching its shell commands, avoid storing raw credentials in its memory, and protect the memory store from unintended writes. The scan is partial: 23 of 24 components were scanned, one file was only partially inspected, and a tool-misuse analyzer was degraded, so the evidence is incomplete.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
supermemory.py
Finding

Imported memory IDs permit JSON writes outside the memory node directory

Content
View full analysis

Impact

An attacker can create attacker-controlled JSON files outside the node directory with the importing process's filesystem privileges. Calling import_from(..., skip_duplicates=False) also permits overwriting existing writable JSON targets. No code execution is established.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (9)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

Invoking Python is a reasonable implementation detail for the declared memory engine. However, execSync executes a shell command containing caller-controlled text. The add and search handlers leave shell substitutions active, and pin interpolates its argument without quoting or validation. This permits actions beyond memory management under the process's privileges. This is an exposed execution capability, not evidence of intentional malicious behavior. The supplied chunk does not include supermemory.py, so the claimed decay, ranking, caching, and import/export functionality cannot be verified.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · memory/nodes/20260825-383.json (reported line 1)May include surrounding context.

json
{"id": "20260825-383", "digest": "stepfun API key 有效(openspace/.env 的 STEPFUN_API_KEY), step-3.5-flash 可正常对话, 大陆直连可用", "keywords": ["st", "te", "ep", "pf", "fu", "un", "ste", "tep"], "weight": 75, "pinned": false, "created_at": "2026-08-25T13:06:34.429833", "last_access": "2026-08-25T13:06:34.429841", "times_accessed": 1, "_auto_weighted": false}

Recommends retaining API keys and passwords indefinitely without a sensitive-data warning.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The guide explicitly recommends pinning API keys and passwords indefinitely in general-purpose agent memory. This increases opportunities for secrets to appear in retrieval results, conversation context, and backups rather than remaining confined to a dedicated credential store.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Activation keywords overlap with ordinary requests unrelated to memory management.

Medium
Category
Trigger Abuse
Confidence
99% confidence
Finding

The trigger list includes generic expressions such as “查一下” (look something up), “总结” (summarize), and “锁定” (lock), without requiring a memory-management context. These can activate the skill for unrelated requests, and no exclusion conditions distinguish ordinary conversation from an intentional memory operation.

Content

No source excerpt is available for this finding.

The skill recommends permanently retaining API credentials without a sensitive-data warning.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The skill explicitly presents API credentials as appropriate for permanent retention and prioritizes pinned memories during retrieval. Treating credentials as routinely retrievable agent context increases disclosure risk through responses, context propagation, and memory exports.

Content

No source excerpt is available for this finding.

Troubleshooting advice redirects the agent around secret redaction.

Medium
Category
Prompt Injection
Confidence
97% confidence
Finding

The stored lesson explicitly recommends switching to a Python heredoc to read a secret after another access method encounters redaction. In an agent memory system, this can resurface as trusted troubleshooting advice and induce circumvention of a protective mechanism, although no actual secret value or exfiltration destination is included.

Content

No source excerpt is available for this finding.

Default execution persists fabricated user memories and a locked behavioral restriction in the real memory store.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Running the script without a recognized command writes invented user preferences and a pinned behavioral restriction into the default ./memory store. Because these records persist and pinned records receive retrieval priority and avoid decay, a demonstration can contaminate subsequent agent context with statements the user never supplied.

Content

No source excerpt is available for this finding.