Back to skill

Security audit

效率中枢

Security checks for vulnerabilities and agentic risk

Overview

This productivity hub is not malicious, but it broadly authorizes memory access, saved chat archives, reminders, backups, web search, and automatic delegation to other skills without clear confirmation boundaries.

Review this skill before installing. It is a convenience hub, but users should only use it where saving chat content, searching prior memory, creating reminders, backing up files, using web search, and invoking other skills are acceptable after explicit user confirmation. It would be safer with narrow triggers, previews before writes/searches/delegation, path limits, retention and deletion controls, and a way to list or cancel scheduled reminders.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

other

Warning
Location
SKILL.md:79
Finding

Unrestricted Access to Persistent Memory, Chat Archives, and Backup Content

Content
View full analysis

Vulnerability Details

File Locations:

  • SKILL.md:79-85
  • SKILL.md:103
  • references/commands.md:8-10
  • references/commands.md:15
  • references/commands.md:59

Vulnerability Type: Unrestricted sensitive Agent data access and persistence
Risk Level: Medium

Relevant Source Segments

The following are faithful English translations of the relevant source instructions:

text
SKILL.md:79-85

User: "Record this: the product should be simple and should not have complex features."
→ Extract key information → Save it to MEMORY.md and record it in the daily log

User: "Find the previous conversation about stock analysis."
→ Search memory/chat-*.md → Return the results
text
SKILL.md:103

→ Aggregate today's conversations → Extract key decisions → Produce a review report
text
references/commands.md:8-10

| "Archive" | Save the conversation to memory/chat-YYYY-MM-DD.md |
| "Save this" | Extract key information into MEMORY.md |
| "Back up X" | Copy X into the backup/ directory |
text
references/commands.md:15

| "Find X" | Select-String "X" memory/chat-*.md |
text
references/commands.md:59

- "Remember my wife's birthday" → Save it to MEMORY.md

Technical Analysis

The Skill instructs the Agent to read historical conversations, extract information from them, write information into persistent memory, and copy user-selected content into a backup directory. The parameter represented by X is not constrained to an approved directory, data type, or sensitivity level.

No requirements are defined for explicit user confirmation, source-path validation, destination-path validation, sensitive-data filtering, retention periods, access controls, or deletion. Consequently, a broad natural-language request can cause private conversation content or other sensitive information to be searched, returned, duplicated, or retained beyond the immediate se ...[truncated 1568 chars]

Remediation
View remediation

Remediation Suggestions

  1. Require explicit confirmation before reading chat archives, writing to MEMORY.md, creating archives, or copying content into a backup directory.
  2. Resolve and validate all source and destination paths against strict allowlisted directories.
  3. Reject absolute paths, parent-directory traversal, symbolic-link escapes, and unsupported file types.
  4. Show the user a preview of the information that will be saved or returned.
  5. Detect and redact credentials, authentication tokens, financial information, health information, and other sensitive personal data by default.
  6. Apply least-privilege file permissions to memory, archive, and backup files.
  7. Define retention and deletion policies for persistent memory, chat archives, and backups.
  8. Record auditable metadata describing what was accessed, where it was stored, and which user approved the operation.
  9. Separate ephemeral task context from long-term memory so ordinary requests do not automatically become persistent records.

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:109
Finding

Ambiguous Requests Can Trigger Overbroad Automatic Skill Delegation

Content
View full analysis

Vulnerability Details

File Locations:

  • SKILL.md:109-114
  • SKILL.md:133-140
  • references/commands.md:42-45

Vulnerability Type: Unconfirmed delegation to tools or other Skills with broader capabilities
Risk Level: Medium

Relevant Source Segments

The following are faithful English translations of the relevant source instructions:

text
SKILL.md:109-114

When the user says "help me do it," "I am too busy," or "I do not have time":
1. Understand the requirement
2. Automatically determine which Skill is needed
3. Invoke the corresponding Skill
4. Return the result
text
SKILL.md:133-140

| Requirement | Skill to invoke |
| Excel processing | automate-excel |
| Stock analysis | stock-analyst-ai-stock-analysis |
| Video generation | seedance-video-gen |
| Decision support | master-decision-system |
| Memory archiving | chat-memory-v2 |
text
references/commands.md:42-45

| "Help me do X" | Automatically determine → Execute |
| "I am too busy" | Automatically determine the most urgent X tasks |
| "I do not have time" | Provide the most efficient solution |
| "Efficiency" | Open the efficiency-hub menu |

Technical Analysis

The Skill treats vague phrases as authorization to infer a task, select another Skill, and execute it. It does not define an allowlist, capability boundary, permission review, argument validation, or confirmation step before delegation.

Delegated Skills may possess materially different capabilities, including file access, persistent-memory writes, web access, content generation, or other tool permissions. Automatically invoking them from an ambiguous request can therefore cross least-privilege boundaries and perform actions that the user did not specifically approve.

The risk is amplified by trigger phrases such as “help me do X,” “I am too busy,” and “I do not have time,” because these phrases do not identify a concrete action ...[truncated 1191 chars]

Remediation
View remediation

Remediation Suggestions

  1. Convert automatic execution into a proposal-and-confirmation workflow.
  2. Before delegation, display the selected Skill, intended action, required permissions, target resources, and expected side effects.
  3. Require explicit confirmation for file access, persistent writes, network requests, scheduling, external-service submission, and additional Skill invocation.
  4. Maintain an allowlist of approved downstream Skills and document the capabilities each one may use.
  5. Grant each delegated Skill only the minimum permissions required for the confirmed operation.
  6. Treat vague trigger phrases as requests for clarification rather than authorization to execute.
  7. Validate and sanitize all arguments passed to downstream Skills.
  8. Prevent delegated Skills from recursively invoking additional Skills unless separately authorized.
  9. Log delegation decisions, permission grants, arguments, and execution outcomes for auditability.
  10. Provide a dry-run mode that describes intended actions without performing side effects.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The description presents the skill as a one-stop handler for vague umbrella requests without defining boundaries for what it should or should not do. This increases the risk that normal user language is interpreted as consent to invoke cross-functional capabilities, including persistence, retrieval, and task scheduling.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The description presents the skill as a one-stop handler for vague umbrella requests without defining boundaries for what it should or should not do. This increases the risk that normal user language is interpreted as consent to invoke cross-functional capabilities, including persistence, retrieval, and task scheduling.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill description and user-facing instructions are entirely in Chinese and appear to assume Chinese-language interaction without offering a language choice. This can violate language or locale policy when a skill imposes a specific language by default rather than allowing user opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The save module encourages archiving conversations, recording ideas, and backing up content, but does not warn users that their conversation data may be written to files or retained for later retrieval. Without clear disclosure and confirmation, users may unknowingly persist sensitive information, creating privacy and data-retention risks.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The command catalog includes saving chat content to local files, copying data to backups, and sending queries to web tools without informing the user that data will be persisted or transmitted externally. In a productivity skill designed for frequent daily use, this lack of transparency creates a meaningful privacy and consent risk, especially when users may share personal schedules, relationships, or work information.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger phrase "帮我做X" is extremely broad and delegates action selection to the skill ("自动判断 → 执行"), creating a high risk of accidental activation and overbroad tool use. In this skill, that danger is amplified because available actions include persistent file writes, backups, reminders, and external web operations, so ordinary conversation could unintentionally cause state changes or data disclosure.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Phrases like "太忙了" and "没时间" are common conversational statements rather than clear commands, so they can trigger the skill unexpectedly during normal dialogue. Because this skill can prioritize tasks and propose or execute follow-on actions in a one-stop workflow context, ambiguous activation increases the chance of unintended reminders, searches, or persistence operations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The search module advertises searching prior chats and files without informing users that historical conversation content may be queried and surfaced. This can expose sensitive prior context unexpectedly, especially when broad triggers or ambiguous requests cause the module to run unintentionally.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The reminder module describes creating reminders and scheduled tasks via cron but does not disclose that it may create persistent scheduled jobs. Users may not understand that the action has ongoing effects beyond the current session, which can lead to unwanted reminders or task execution behavior.

Content

No source excerpt is available for this finding.