Back to skill

Security audit

GAN Evolution Engine

Security checks for vulnerabilities and agentic risk

Overview

This skill performs GAN-style skill evolution, but it runs and writes untrusted variant code without real isolation and mishandles credentials, so it needs Review before installation.

Do not run this on a real workspace or with real credentials unless it is fixed and reviewed. If testing is necessary, use a disposable container or VM with no valuable files, no inherited API keys, and network access disabled; remove or isolate bundled evolved skill outputs and hooks; rotate the exposed EvoMap secret; redact credential logging; validate all generated paths; and pin dependencies.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (7)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/discriminator.py:86
Finding

Unsandboxed Execution of Target-Controlled Benchmark and Runtime Scripts

Content
View full analysis
float: """Run skill's benchmark suite and return fitness score (0-1).""" benchmark_script = skill_path / "scripts" / "benchmark.py" if not benchmark_script.exists(): # No benchmark → use simple correctness check return self._simple_correctness_check(skill_path) try: # Run benchmark in skill's workspace context result = subprocess.run( ["python3", str(benchmark_script)], cwd=skill_path, capture_output=True, text=True, timeout=60 ) ``` Additional execution paths invoke a target-controlled `scripts/run.py`: ```python subprocess.run( ["python3", str(main_script), "--benchmark"], capture_output=True, timeout=30 ) ``` ```python process = subprocess.Popen( ["python3", str(main_script), "--benchmark"], stdout=subprocess.PIPE, stderr=subprocess.PIPE ) ``` ### Technical Analysis The discriminator treats `scripts/benchmark.py` and `scripts/run.py` from the selected skill as trusted Python programs. These programs run with the same operating-system identity, environment variables, filesystem access, and network access as the evolution engine. The configured timeouts only constrain execution duration. They do not prevent filesystem modification, credential access, network connections, child-process creation, or attacks against other local processes. No container, namespace, restricted user, syscall filter, read-only mount, network policy, or environment-variable filtering is applied. This also contradicts the safety claims in `SKILL.md:169-175`, which state that variants are sandboxed and have no networ ...[truncated 1202 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/population.py:33
Finding

Model-Controlled Path Traversal Enables Arbitrary File Writes

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:39
Finding

Hard-Coded EvoMap Credentials Distributed in Setup Instructions

Content
View full analysis
~/.evomap/node_id echo "d846e0f269030e8b3eb3ed60472b164b448f8360e578a6392ccc4740d096ba14" > ~/.evomap/node_secret ``` ### Technical Analysis The installation instructions contain a concrete node identifier and secret-shaped value. They instruct every user to save these values in plaintext under `~/.evomap`. Secrets included in a distributed project must be considered compromised because every recipient can retrieve them. The instructions also do not set restrictive file permissions, so the resulting files may be readable beyond the intended account depending on the user's umask and host configuration. ### Attack Path 1. An attacker obtains the publicly or internally distributed project. 2. The attacker reads the embedded node identifier and secret from `SKILL.md`. 3. If the credential remains active, the attacker uses it with the EvoMap publishing interface. 4. Requests appear to originate from the shared node identity. 5. Separately, users following the instructions persist the same credential in plaintext, increasing the number of exposed copies. ### Impact Assessment If the embedded value is valid, an attacker may impersonate the associated EvoMap node and perform any publishing operations authorized for that identity. This can cause unauthorized submissions, loss of attribution integrity, abuse of service quotas, or account revocation. Even if the value is expired or demonstrative, distributing realistic secret material establishes an unsafe credential-management pattern. ]]>
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/gan_evolution.py:191
Finding

EvoMap Secret Is Exposed in Logs and Process Arguments

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
evolved_full/self-improving-agent-v3/hooks/openclaw/handler.js:9
Finding

Persistent Bootstrap Instruction Injection Can Poison Agent Memory

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
evolved_full/self-improving-agent-v3/scripts/benchmark.py:24
Finding

Bundled Benchmark Writes to a Hard-Coded External Workspace

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
requirements.txt:1
Finding

Unpinned and Incomplete Python Dependency Specification

Content
View full analysis
=1.0.0 psutil>=5.9.0 ``` Additionally, `scripts/population.py:20` imports `filelock`, but `filelock` is not declared in `requirements.txt`: ```python import filelock ``` ### Technical Analysis The dependency constraints use open-ended lower bounds. A future installation may therefore resolve package versions that were never reviewed or tested with this project. No lockfile or package hashes are present in the audited tree. The project also imports an undeclared third-party package. This encourages ad hoc installation and makes it harder to reproduce or audit the actual runtime environment. The audit did not confirm a currently malicious dependency. The issue is the unsafe dependency-management practice and the resulting supply-chain and reproducibility risk. ### Attack Path 1. A user installs dependencies at a later date. 2. The package manager resolves the newest versions satisfying the broad lower bounds. 3. Newly introduced or compromised package code is installed without being covered by the original review. 4. The dependency executes with the installing user's privileges during installation or runtime. 5. Alternatively, the missing `filelock` dependency causes users or automation to install an arbitrary version outside the reviewed manifest. ### Impact Assessment Potential impact depends on the behavior of future resolved releases. A compromised dependency would run with the same privileges as the evolution engine and could access source code, API keys, network resources, and user-writable files. Even without compromise, incompatible releases can break security assumptions and produce non-reproducible behavior. ]]>
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (159)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The file presents itself as a GAN-based evolution engine, but the static findings indicate materially different behavior such as undeclared hooks, bootstrap context injection, filesystem writes, benchmarking, or post-tool monitoring. A description/behavior mismatch is dangerous because it can conceal privileged actions from reviewers and users, undermining informed consent and making malicious or risky side effects harder to detect.

Content

No source excerpt is available for this finding.