Back to skill

Security audit

麦赛尔夫

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent investment-research agent template with disclosed workspace logging and report creation, with no hidden exfiltration, destructive behavior, or credential use found.

Install only in a dedicated workspace, review changes to MEMORY.md and SKILLS.md before letting them persist, and pin or vet pdfminer.six before using the optional PDF parsing workflow. Do not provide brokerage credentials or login-gated data to this agent.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:288
Finding

Unpinned Third-Party Dependency Installation

Content
View full analysis

Vulnerability Details

File Locations:

  • SKILL.md:288-292
  • references/methodology.md:108-117

Vulnerability Type: Unpinned Python dependency installation
Risk Level: Medium

Vulnerable Code:

SKILL.md:288-292:

bash
pip install pdfminer.six
python -m venv venv
source venv/bin/activate
pip install pdfminer.six

references/methodology.md:108-117:

bash
pip install pdfminer.six
python -m venv venv
source venv/bin/activate
pip install pdfminer.six

Technical Analysis

The Skill instructs users to install pdfminer.six without specifying an audited version, validating package hashes, or using a locked dependency manifest. Consequently, the installed package is whichever release the package index resolves at execution time, rather than the version reviewed with this Skill.

Python package installation may execute package build or installation logic. If the package, one of its transitive dependencies, the configured package index, or the dependency distribution channel is compromised, following these instructions could execute attacker-controlled code locally.

Creating a virtual environment reduces interference with system-level Python packages but does not provide a security sandbox. Code installed or run inside the environment generally retains the invoking user's access to files, network resources, environment variables, and other resources available to that user.

No evidence indicates that pdfminer.six is currently malicious. The vulnerability is the unsafe, non-reproducible dependency-installation practice and the resulting supply-chain exposure.

Attack Path

  1. An attacker compromises a future release of pdfminer.six, a transitive dependency, an account authorized to publish it, or a package index used by the victim.
  2. A user follows the Skill's unpinned pip install pdfminer.six instruction.
  3. pip resolves and downloads the attacker-controlled o ...[truncated 1098 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin pdfminer.six to a specifically reviewed version rather than installing the latest available release:

    bash
    python -m pip install "pdfminer.six==REVIEWED_VERSION"
    
  2. Generate a lockfile containing cryptographic hashes and require hash verification during installation:

    bash
    python -m pip install --require-hashes -r requirements.txt
    
  3. Pin and hash all transitive dependencies, not only the top-level package.

  4. Document the trusted package index explicitly and prevent unintended fallback to untrusted or organization-internal indexes.

  5. Retain the virtual-environment recommendation, but clarify that it provides dependency isolation rather than a security boundary.

  6. Run PDF parsing in a restricted environment with only the required input and output directories mounted. Avoid exposing credentials or unrelated workspace files to the parser.

  7. Review dependency provenance, release signatures where available, vulnerability advisories, and the resolved dependency tree before updating the pinned version.

  8. Replace the duplicate installation instructions in both affected files with one centrally maintained, reproducible setup procedure.

Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (8)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill explicitly requires writing a report file and updating a daily memory log, but it does so as a mandatory workflow without any user-facing warning or consent boundary around workspace modification. In an agent setting, silent file creation/modification can overwrite existing work, create persistent artifacts, or normalize unattended state changes that later tasks may rely on.

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 the agent to routinely update persistent instruction files such as MEMORY.md and SKILLS.md, including after receiving new standing instructions or after an error/redo. Allowing an agent to modify its own long-lived policy/process files creates a durable prompt-injection surface where adversarial content from a task can be written into future session context and silently change behavior over time.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction 'all English' imposes a specific language requirement in natural language. This can violate language/locale policy because the file does not offer the user a language choice or explain a justified region-specific constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instruction 'File and path names must be full English' imposes a language/locale restriction as a hard rule. This is a natural-language policy concern because it forbids other scripts without offering user choice or documenting a clear region-specific need.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instruction requires all report file paths and filenames to be 'full English' and explicitly bans CJK characters. This is a natural-language locale restriction, and the file does not provide user choice, opt-in, or a region/compliance justification for enforcing that language constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The file directs that the report body must be written in the principal's language, which is a locale/language constraint. While potentially reasonable, it does not explicitly offer the user a language choice or describe the constraint as optional or justified, so it fits the policy-violation category.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The template includes a dedicated 'Language' field with examples like 'Chinese, English', which can encourage a skill configuration that fixes a single language preference. Because there is no accompanying note that language choice is optional or user-adjustable, this may conflict with the policy against forcing a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The document states an organizational rule of 'English file paths only' at L011, and then includes Chinese comments in the installation example at L116-L117. This introduces a locale-specific instruction in the skill content without offering a language choice or documenting why non-English text is required.

Content

No source excerpt is available for this finding.