Back to skill

Security audit

meta-native-autonomous-discovery

Security checks for vulnerabilities and agentic risk

Overview

The skill is not clearly malicious, but it overstates its capabilities and asks the agent to persist and reuse user/task notes, including potentially rewriting its own instructions.

Install only if you are comfortable with a skill that keeps local cross-session notes about usage, errors, and preferences. Do not let it automatically rewrite SKILL.md, and review or disable learned_patterns.json storage if task content may include sensitive information.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • 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
Findings (1)

T02 · Agent Memory Poisoning

Warning
Location
scripts/learner.py:10
Finding

Persistent Agent Memory Poisoning Through Unrestricted Stored Notes

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The implementation does not exhibit the sophisticated meta-agent behaviors described. It only records invocation statistics and notes in a local JSON file (learned_patterns.json). While this could be a supporting component for a larger meta-evolver loop, the supplied code chunk itself does not perform self-validation, reflection, orchestration, or autonomous improvement. Therefore the declared description materially overstates the actual behavior of this code.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill documents file-writing behavior via persistent memory (learned_patterns.json) and even suggests writing experience back into SKILL.md, but it declares no permissions or allowed-tools scope. That mismatch weakens operator visibility and policy enforcement, increasing the risk of unexpected local file modification or persistence beyond what a caller intended.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill is entirely presented in Chinese and includes examples that set 输出语言 to 中文, but it does not offer an opt-in or alternative language choice. Under the policy, forcing a specific language without user choice is a natural-language policy violation unless clearly justified as region-specific, which is not documented here.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill is presented as an autonomous-discovery enhancer, but the latter section broadens it into a general self-evolving memory and adaptation system. That functional expansion increases the attack and privacy surface because a caller may invoke the skill expecting bounded task support, not cross-session behavioral accumulation and file mutation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill introduces persistent storage of usage history, error patterns, and user preferences that are not necessary for basic autonomous discovery. This expands data collection scope beyond the stated function, creating privacy and retention risk if sensitive task history or preference data is stored without minimization or consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill states that user preferences and usage-derived data will be stored persistently, but it does not provide an explicit privacy warning, consent flow, or data handling boundaries. Silent persistence of preference/history data can expose sensitive behavioral information and violate user expectations or organizational data policies.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly directs persistent collection of user preferences, usage frequencies, error patterns, and notes across sessions in a local memory file. Cross-session retention of behavioral and preference data creates a meaningful privacy and data-governance risk, especially because the notes/examples could contain sensitive task content and there is no retention, minimization, or access-control guidance.

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 writing learned experience back into SKILL.md, effectively modifying its own operational instructions. Self-modifying skill definitions can undermine reviewability and trust boundaries, allowing drift, accidental corruption, or injection of unsafe guidance into future runs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The report is entirely written in Chinese, and there is no indication that language selection is optional or that the skill is region-specific. Under the policy for natural-language violations, forcing a specific language without user opt-in is in scope.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

Line L52 claims the learning module works automatically after each use and requires no manual maintenance. But lines L57-L69 explicitly tell the operator to run python scripts/learner.py ... commands after use and to provide fields like --capability, --error, and --note, which contradicts the claim of automatic, maintenance-free operation.

Content

No source excerpt is available for this finding.