Back to skill

Security audit

glr-adapter-builder

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent GLR adapter scaffolding package with disclosed build, test, research, and loader-staging workflows, though users should run its package checks in a clean environment.

Install only if you are building authorized GLR adapter projects. Use a clean shell or CI job without unrelated secrets before running generated package checks, verify any referenced glr-cli workflow separately, and keep loader/live-runtime work limited to explicitly permitted offline or test targets.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • 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
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as an adapter-builder for GLR environments, but it also directs broader software packaging, environment setup, dependency validation, and test execution workflows. This mismatch can cause operators or policy layers to underestimate the operational reach of the skill, leading to execution of build/install/test actions in contexts where only design or refactoring assistance was expected.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 17)May include surrounding context.

md
`SKILL.md`; all `scripts/`, `assets/`, and `references/` paths below are

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
73% confidence
Finding

The script copies nearly the entire parent process environment into child build, install, and test processes, excluding only PYTHONPATH and PYTHONHOME. In a CI or developer environment this can expose sensitive tokens, credentials, proxy settings, or internal endpoints to untrusted build backends, package install hooks, or test code, all of which are executed by this workflow.

Content

Scanner excerpt · assets/templates/check_wheel.template.py (reported line 21)May include surrounding context.

python
project = Path(__file__).resolve().parents[1]
    environment = {
        key: value
        for key, value in os.environ.items()
        if key.upper() not in {"PYTHONPATH", "PYTHONHOME"}
    }
    with tempfile.TemporaryDirectory(prefix="glr-wheel-check-") as directory:

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill describes commands that can read/write files, invoke shell commands, inspect environment state, and access networked sources, but it does not declare an explicit tool/permission scope. That creates an authorization ambiguity where an agent or orchestrator may grant broader capabilities than intended, increasing the risk of unsafe execution in a high-impact environment that touches game runtimes, packages, and external research sources.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · assets/templates/check_wheel.template.py (reported line 27)May include surrounding context.

python
with tempfile.TemporaryDirectory(prefix="glr-wheel-check-") as directory:
        work = Path(directory).resolve()
        distributions = work / "dist"
        subprocess.run(
            [
                sys.executable,
                "-I",

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · assets/templates/check_wheel.template.py (reported line 52)May include surrounding context.

python
install = [str(python), "-I", "-m", "pip", "--isolated", "install"]
        if args.find_links is not None:
            install.extend(["--find-links", str(args.find_links.resolve())])
        subprocess.run(
            [*install, str(wheels[0]), "pytest>=8.3"],
            cwd=work,
            env=environment,

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · assets/templates/check_wheel.template.py (reported line 58)May include surrounding context.

python
env=environment,
            check=True,
        )
        subprocess.run(
            [str(python), "-I", "-m", "pip", "--isolated", "check"],
            cwd=work,
            env=environment,

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · assets/templates/check_wheel.template.py (reported line 64)May include surrounding context.

python
env=environment,
            check=True,
        )
        subprocess.run(
            [
                str(python),
                "-I",

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · assets/templates/check_wheel.template.py (reported line 79)May include surrounding context.

python
env=environment,
            check=True,
        )
        subprocess.run(
            [
                str(python),
                "-I",

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

The manifest-level intent emphasizes preserving privacy, and several later sections prohibit persisting private traces and identifiers. However, this section explicitly directs the user to record URLs, publisher, access time, and update metadata in a research manifest. While this is likely aimed at public sources, the blanket privacy framing creates a mild contradiction in the documentation about what metadata should be retained.

Content

No source excerpt is available for this finding.