Back to skill

Security audit

Life Memory Logger - 生活记忆记录器

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed relationship-memory helper, but it asks the agent to persist and index sensitive personal details with limited containment guidance.

Review before installing. Use this only if you are comfortable keeping relationship, health, event, and follow-up details in local memory files and indexes. Avoid recording sensitive third-party facts unless you have clear consent, periodically review and delete stored records, and prefer implementations that use safe generated IDs, restricted file permissions, and complete deletion across indexes and reminders.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:81
Finding

Plaintext Persistent Storage of Sensitive Personal Data

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 81-116 and 163-172
Vulnerability Type: Plaintext storage of sensitive personal and third-party information
Risk Level: High

Vulnerable Documentation

The Skill prescribes persistent JSON storage using the following structure:

text
/memory/life-memories/people/[person-name].json
/memory/life-memories/upcoming-events.json
/memory/life-memories/follow-ups.json
/memory/life-memories/master-index.json

The documented person record includes identifying information, preferences, allergies, birthdays, conversation summaries, and action items. The privacy rules also explicitly permit storage of sensitive health, financial, relationship-conflict, and third-party information after confirmation.

Technical Analysis

The Skill directs the Agent to create persistent personal profiles in ordinary JSON files. No encryption at rest, restrictive file permissions, access-control checks, data-retention limits, secure deletion procedures, or isolation from other Skills and local processes are specified.

The master index, event database, follow-up database, and individual person records can duplicate or reference the same sensitive information. Consequently, deleting a fact from one person record may not remove it from indexes, summaries, backups, or generated reminders.

User confirmation does not mitigate the technical exposure created after storage. The risk is particularly significant because the collected information may concern people who never interacted with or consented to the Agent.

Attack Path

  1. A user provides a conversation containing a birthday, allergy, health condition, financial detail, relationship conflict, or other private fact.
  2. The Skill instructs the Agent to extract that fact and associate it with an identified person.
  3. The Agent writes the information to a plaintext JSON record and may add references to the master index, e ...[truncated 870 chars]
Remediation
View remediation

Remediation Suggestions

  • Do not store health, financial, relationship-conflict, authentication, or third-party private information by default.
  • Require explicit, granular consent for each sensitive record rather than relying on general confirmation.
  • Encrypt sensitive records at rest using keys stored outside the memory directory.
  • Restrict files and directories to the minimum required operating-system identity, such as owner-only permissions.
  • Separate indexes from sensitive record contents and avoid duplicating private facts in summaries or reminder files.
  • Establish configurable retention periods and automatically purge expired records.
  • Implement deletion across person records, indexes, reminders, cached summaries, and backups.
  • Record provenance and consent status for every stored fact.
  • Provide an inventory and export mechanism so users can review all retained information.
  • Add tests verifying that sensitive data is neither logged nor exposed through generated reminders.

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:81
Finding

Unsanitized Person Name Used as a Persistent File Path Component

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 81
Vulnerability Type: Potential path traversal through conversation-derived filenames
Risk Level: High

Vulnerable Documentation

The documented file-mapping rule is:

text
/memory/life-memories/people/[person-name].json

Technical Analysis

The Skill maps a person's name, which originates from conversational input, directly to a filename. It does not require filename normalization, path-separator rejection, canonical-path validation, generated identifiers, or containment checks.

In a literal implementation, a crafted name containing traversal sequences, path separators, an absolute path, or platform-specific filename syntax could cause file operations to escape the intended people directory. Appending a JSON extension alone does not prevent traversal.

The package contains no implementation scripts, so the existence of an exploitable file-writing implementation cannot be verified. Nevertheless, the prescribed storage design is unsafe and can lead implementers or Agents to perform attacker-influenced file access without validation.

Attack Path

  1. An attacker supplies or causes the user to repeat a crafted person name containing directory traversal components.
  2. The Agent interprets the value as the person identifier to be remembered.
  3. A literal implementation substitutes the value into the documented people-file path.
  4. The filesystem resolves the traversal components before the file is opened.
  5. The Agent writes to or reads from a JSON file outside the intended memory directory.
  6. Depending on the Agent's filesystem permissions and application behavior, the operation may overwrite application data, alter another JSON configuration, or disclose an existing file.

Impact Assessment

The maximum scope is limited by the filesystem privileges of the Agent process. Within those privileges, successful exploitation could permit ...[truncated 354 chars]

Remediation
View remediation

Remediation Suggestions

  • Never use a display name directly as a filename.
  • Generate a random UUID or cryptographically strong opaque identifier for each person record.
  • Store the human-readable name only inside the JSON document.
  • If filename-derived identifiers are unavoidable, apply a strict allowlist and reject path separators, traversal tokens, control characters, drive prefixes, and reserved filenames.
  • Resolve the canonical storage root and canonical target path before every operation.
  • Verify that the target remains a descendant of the canonical storage root.
  • Open files using APIs that prevent symbolic-link traversal where supported.
  • Use exclusive creation and safe-update patterns to reduce overwrite and race-condition risks.
  • Add tests covering traversal sequences, absolute paths, alternate separators, encoded separators, symbolic links, and platform-specific reserved names.
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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (4)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README describes collecting and acting on sensitive interpersonal memory data such as birthdays, preferences, and promises, but provides no warning about privacy, consent, retention, or reminder side effects. This can lead users to store personal data without understanding the sensitivity or the need to obtain permission, increasing the risk of privacy harm or misuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly stores sensitive personal data such as health information, relationship details, birthdays, and conversation summaries in local JSON files and aggregate indexes, but it does not present a clear upfront warning about retention risk, local file exposure, indexing, or reminder propagation before collection begins. Even though later sections mention ethics and consent, the design encourages accumulation of sensitive third-party data, which increases harm if the host device, memory directory, or reminder outputs are accessed by others.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The title, description, and all usage examples are written only in Chinese, which can imply a fixed language requirement. The file does not state that Chinese is optional, nor does it justify a Chinese-only scope as a region-specific tool.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill description, examples, prompts, and operating instructions are entirely in Chinese, and no alternative language option or opt-in is described. Under the policy, forcing a specific language without user choice can be a natural-language policy violation unless the locale constraint is clearly documented and justified.

Content

No source excerpt is available for this finding.