Back to skill

Security audit

long-text-summarizer

Security checks for vulnerabilities and agentic risk

Overview

The summarizer works as advertised, but it also includes a persistent self-learning tracker that can store usage notes and preferences beyond the immediate summary task.

Review this carefully before installing. The summarization and chunking behavior is local and purpose-aligned, but the self-learning script can create persistent local records of usage, errors, notes, and preferences. Avoid using the learner component with sensitive documents or confidential workflow details unless you are comfortable with that plaintext local history and have a way to manage or delete it.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The stated purpose is long-text summarization, but the skill also introduces operational history tracking, failure analytics, user-preference persistence, and self-improvement state in learned_patterns.json. This is dangerous because it expands the trust boundary beyond summarization into collection and storage of behavioral data, creating privacy, retention, and misuse risks that users would not reasonably expect from the description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file explicitly describes and implements a generic cross-skill learning/tracking module rather than functionality specific to long-text summarization. In this skill context, bundling a reusable telemetry-style component that can log usage, errors, notes, and preferences across arbitrary skills expands data collection scope beyond user expectations and increases privacy risk and attack surface.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code loads and writes a learned_patterns.json file containing arbitrary capability names, user preferences, operational history, error types, and free-form notes for any skill directory. For a summarization skill, this is unjustified persistent behavior that can capture sensitive user data or internal workflow details unrelated to the advertised purpose.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
87% confidence
Finding

The skill documents commands that write files (for example chunks.json and the learner subsystem’s persisted records) but declares no explicit tool scope or permissions. In an agent environment, undocumented write capability can let a summarization skill modify local state unexpectedly, which weakens least-privilege guarantees and makes abuse or accidental persistence harder to control.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The self-evolution and usage-tracking subsystem is unrelated to the core summarization function and adds persistent state based on prior interactions. Even if intended for quality improvement, such behavior can accumulate sensitive information about documents, failures, or user habits, increasing privacy exposure and making the skill more dangerous than a stateless summarizer.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Recording operational history and analyzing failures is unjustified for a summarization skill because it enables retention of interaction metadata without a clear need tied to the user-requested task. In the context of summarizing long documents, failure notes and usage logs could inadvertently capture sensitive document topics, filenames, or user behavior patterns.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The module docstring and command-line help/user messages are written in Chinese only, which imposes a specific language on users without any opt-in or alternative locale. The stated policy flags language/locale constraints when the skill does not offer a language choice or clearly justify the restriction.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

record_op stores timestamps, capability names, success/failure state, error labels, and free-form notes in a local JSON file, with recent history retained automatically. Free-form notes and error details can easily contain sensitive prompts, document metadata, or operational context, and the code provides no warning, sanitization, or retention controls.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

record_preference persistently stores key/value preferences to disk without any visible disclosure, consent flow, or sensitivity checks. Preference values may include personal or confidential settings, so silent persistence can violate user expectations and create privacy/compliance issues if the host environment is shared or compromised.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This code writes JSON results to a user-specified path when --out is provided. Although the operation is visible in code, there is no confirmation prompt or explicit warning comment/docstring about modifying or potentially overwriting an existing file at that path.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The module description, usage guidance, and status messages are presented in Chinese only, with no indication of language choice or locale scope. That can violate language/locale policy when users are not given an explicit opt-in or alternative language.

Content

No source excerpt is available for this finding.