Back to skill

Security audit

SeekDB Memory Setup

Security checks for vulnerabilities and agentic risk

Overview

This cloud-memory setup skill is purpose-aligned, but it enables automatic remote memory capture and persistent plugin configuration through risky install and command patterns.

Install only if you trust the m0 service, endpoint, package source, and publisher with conversation-derived memory. Review or pin the exact packages first, avoid entering secrets while autoCapture is enabled, confirm where memories are stored and how they can be deleted, and prefer disabling automatic capture until you have clear controls.

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 (3)

T08 · Insecure Dependencies

Error
Location
SKILL.md:119
Finding

Unpinned Third-Party Package Installation

Content
View full analysis
Remediation
View remediation
clawhub install m0@ ``` 2. Verify package integrity using publisher signatures, registry provenance attestations, or trusted SHA-256 checksums before execution. 3. Explicitly configure an approved registry instead of relying on ambient package-manager configuration. 4. Prefer a local, isolated installation over a global installation. 5. Review the exact `m0` package contents before adding the package to the gateway load path. 6. Use a lockfile or immutable package artifact so repeated installations retrieve identical content. 7. Run the plugin under a least-privileged account with restricted filesystem and network access. ]]>

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:23
Finding

Command Injection Through Unsanitized Host Command Discovery

Content
View full analysis
/dev/null | grep '\.json' | xargs dirname | sed "s|~|$HOME|") ``` ### Technical Analysis `{CLAW_CMD}` originates from a local directory name or, when no candidate is detected, potentially from user-supplied text. The workflow does not validate the value as a safe executable name before placing it directly into a shell command. Shell metacharacters, whitespace, command substitutions, or redirection operators in the substituted value can alter the intended command structure. This creates a command-injection boundary between an untrusted identifier and the shell interpreter. A malicious local directory and matching JSON file can influence automatic discovery. The file only needs to satisfy the documented checks for the strings `"gateway"` and `"plugins"`. If the resulting name is selected and directly substituted into the command template, attacker-controlled shell syntax may be evaluated. ### Attack Path 1. An attacker who can write within the user's home directory creates a crafted hidden directory whose name contains shell syntax. 2. The attacker creates the expected matching JSON file inside that directory. 3. The file includes the strings `"gateway"` and `"plugins"` so that it passes the discovery checks. 4. The Skill reports the crafted directory name as a candidate and assigns it to `{CLAW_CMD}`, either automatically when it is the only candidate or after unsafe user input. 5. The age ...[truncated 656 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:173
Finding

JavaScript Injection Through Unescaped Configuration Values

Content
View full analysis
/dev/null | grep '\.json' | xargs dirname | sed "s|~|$HOME|") node -e " const fs = require('fs'); const clawRoot = '${CLAW_ROOT}'; const cfgPath = clawRoot + '/{CLAW_CMD}.json'; const cfg = JSON.parse(fs.readFileSync(cfgPath, 'utf8')); if (!cfg.hooks) cfg.hooks = {}; if (!cfg.hooks.internal) cfg.hooks.internal = {}; cfg.hooks.internal.enabled = true; if (!cfg.plugins) cfg.plugins = {}; cfg.plugins.enabled = true; if (!cfg.plugins.allow) cfg.plugins.allow = []; if (!cfg.plugins.allow.includes('m0')) cfg.plugins.allow.push('m0'); if (!cfg.plugins.load) cfg.plugins.load = {}; if (!cfg.plugins.load.paths) cfg.plugins.load.paths = []; const pluginPath = clawRoot + '/extensions/m0'; if (!cfg.plugins.load.paths.includes(pluginPath)) cfg.plugins.load.paths.push(pluginPath); if (!cfg.plugins.slots) cfg.plugins.slots = {}; cfg.plugins.slots.memory = 'm0'; if (!cfg.plugins.entries) cfg.plugins.entries = {}; cfg.plugins.entries['m0'] = { enabled: true, config: { apiKey: '{AK}', baseUrl: '{ENDPOINT}', autoCapture: true, autoRecall: true, recallLimit: 10 } }; fs.writeFileSync(cfgPath, JSON.stringify(cfg, null, 2) + '\n'); console.log('Config written. Gateway will auto-restart.'); " ``` ### Technical Analysis The values `{CLAW_ROOT}`, `{CLAW_CMD}`, `{AK}`, and `{ENDPOINT}` are interpolated directly into a JavaScript program passed to `node -e`. They are embedded inside JavaScript single-quoted string literals without escaping. A value containing a single quote and valid JavaScript syntax can terminate the intended string and inject additional statements. Because the entire JavaScript program is itself enclosed in a shell double-quoted argument, shell-sensitive substitutions may also be evaluat ...[truncated 1521 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (11)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill describes cloud memory installation but does not clearly warn that conversation content will be automatically uploaded to a remote service for persistent storage. Users may provide sensitive information without understanding that it will leave the local environment and be retained externally.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill's stated purpose is to enable automatic injection and persistent write-back of conversational memory to a cloud backend. This creates continuous remote retention of user content, which can capture sensitive data by default and materially changes the privacy model of the host tool.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 165)May include surrounding context.

  1. 运行 {CLAW_CMD} ltm health 确认插件连接正常
  2. 运行 {CLAW_CMD} plugins list 确认 m0 已加载并分配到 memory slot
  3. 使用 message 工具发送安装完成的确认消息,必须指定 channel="{CHANNEL}" 和 target="{TARGET}",内容包含验证结果
  4. 删除此文件:rm ${CLAW_ROOT}/workspace/BOOT.md
text

### 3c. 原子配置写入

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The configuration explicitly sets autoCapture and autoRecall to true with no filtering or approval step, meaning conversation-derived content may be automatically stored remotely and later resurfaced. This can ingest secrets, personal data, or other sensitive context without the user making a deliberate per-item decision.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation trigger includes broad natural-language phrases such as '配置云端记忆', '安装记忆插件', and 'setup memory', which can plausibly appear in ordinary discussion and may cause the skill to activate without the user intending a system-changing installation workflow. Because the skill performs package installation, configuration edits, and network calls, accidental activation raises the risk materially.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · SKILL.md (reported line 79)May include surrounding context.

1a. 确认服务正常

bash
curl -s "{ENDPOINT}/health"

返回 {"status":"ok"} 说明服务运行正常,继续。否则告知用户先启动服务。

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill instructs the agent to create a BOOT.md file that, after restart, uses a message tool to send status information to an external channel. That outbound messaging capability is not necessary to install the plugin and expands the skill's authority beyond local setup into external communications, creating a data exfiltration and unauthorized notification risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest context says the skill should be used to install or set up cloud memory when the user provides an access key or asks to configure memory. However, the documented behavior in this file does not perform installation or plugin configuration; it only retrieves a public IP, tells the user to contact an administrator, and checks a health endpoint after whitelisting.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The surrounding manifest context states the skill should run when a user provides an Access Key beginning with 'ak_'. But the file contains no step that accepts, validates, or applies an access key; instead it focuses entirely on public-IP discovery and whitelist approval.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs the agent/user to query third-party IP-discovery services without warning that this discloses the user's IP and request metadata to external providers. In a setup skill, users may reasonably expect local configuration steps, so silently causing outbound requests to unrelated services creates an avoidable privacy and data-exposure risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill tells the user to send their public IP to an administrator for whitelisting but provides no privacy or retention notice. While operationally plausible, this encourages sharing network-identifying information without informing the user how it will be used, stored, or protected, which increases privacy and trust risk.

Content

No source excerpt is available for this finding.