Back to skill

Security audit

meta-agent-browser-core

Security checks for vulnerabilities and agentic risk

Overview

The skill is not overtly malicious, but it promotes broad self-evolving agent behavior and persistent memory without enough boundaries or user control.

Install only if you are comfortable with a skill that asks the agent to participate in persistent self-learning workflows. Keep its use narrow, review or clear learned_patterns.json regularly, and do not let stored notes become trusted instructions without human review.

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:12
Finding

Unvalidated Notes Can Poison Persistent Agent Learning State

Content
View full analysis

Vulnerability Details

File Location: scripts/learner.py:12-16
Related Instructions: SKILL.md:19, SKILL.md:22-24, SKILL.md:34
Vulnerability Type: Persistent storage of untrusted agent-learning content
Risk Level: Medium

Vulnerable Code

python
def record(op_ok=True, note=""):
    d = load(); d["total_ops"] += 1
    if not op_ok:
        d["total_fails"] += 1
    if note:
        d.setdefault("patterns", []).append(note)
    json.dump(d, open(MEM, "w", encoding="utf-8"), ensure_ascii=False, indent=2)
    return d

The command-line interface supplies the untrusted value directly to this function:

python
rec = (sys.argv[2] if len(sys.argv) > 2 else "")
print(json.dumps(record(ok, rec), ensure_ascii=False))

Technical Analysis

The record() function appends the supplied note verbatim to the persistent patterns array in scripts/learned_patterns.json. It applies no schema validation, provenance tracking, trust classification, size limit, content filtering, or approval process.

The skill documentation directs the agent to record failure patterns and inject the learner into a cross-session evolution loop. Consequently, task content, browser-derived content, or model-generated reflections may be promoted into persistent learning state. If a later learner or meta-evolver component interprets these entries as behavioral guidance, instruction-like attacker content can continue influencing the agent after the original task has ended.

The reviewed package does not contain a component that reads patterns and executes it as code or instructions. Therefore, direct code execution is not established. The security risk arises from the documented intended reuse of this persistent state by external agent-learning components.

Attack Path

  1. An attacker places adversarial instructions in content processed during a skill task, or otherwise influences the reflection note passed to learner.py.
  2. The agent or an orch ...[truncated 1133 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace free-form persistent notes with a strict structured schema containing fields such as event type, trusted source, timestamp, task identifier, and bounded evidence.
  2. Treat all task-derived, browser-derived, and model-generated text as untrusted observations rather than executable rules or behavioral instructions.
  3. Enforce maximum lengths, entry-count limits, and retention periods to prevent persistent storage exhaustion.
  4. Separate raw observations from approved learning rules. Require explicit human review or a trusted validation stage before promoting an observation into reusable behavioral memory.
  5. Reject or quarantine entries containing instruction-like directives, role changes, tool requests, or attempts to override system and safety policies.
  6. Ensure downstream components never concatenate stored notes into privileged prompts without clear untrusted-data delimiters and explicit instructions not to follow embedded directives.
  7. Record provenance and integrity metadata so downstream consumers can distinguish trusted system-generated rules from externally influenced content.
  8. Provide an administrative mechanism to inspect, revoke, and reset learned state.
  9. Use atomic file replacement and restrictive file permissions when writing the JSON state to reduce corruption and unauthorized local modification risks.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (5)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill markets itself as a powerful self-verifying, self-evolving meta-agent that can be directly invoked, but the visible artifact provides only high-level claims without concrete implementation or guardrails. This kind of capability inflation is dangerous because downstream agents or users may over-trust the skill, grant it broader autonomy, or rely on nonexistent safety checks such as self-verification and reflection.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The natural-language description and instructions are presented entirely in Chinese, with no indication that users may choose another language or that the locale is intentionally region-specific. This can violate language/locale policy when a skill implicitly enforces a language without user opt-in.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The invocation guidance says to directly use the skill for the entire agent-browser-core domain, which is overly broad and lacks task, context, or safety preconditions. Broad trigger language increases the chance of inappropriate automatic activation, privilege creep, and use outside intended boundaries, especially for a skill claiming orchestration and persistent learning behaviors.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The markdown content is entirely written in Chinese, including headings and operational descriptions, with no indication that language selection is optional or that the skill is region-specific. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.