ClawPolicy

• Outcome: Review
• Latest audit: Sep 11, 2026 · 3:27 PM UTC
• Version: 3.0.2

Overview

The skill is coherent, but users should review it because it installs and runs an unpinned PyPI policy-engine package that can create local agent policy files.

Install only if you are comfortable running the current PyPI `clawpolicy` package under your user account. Prefer pinning the intended version, for example `clawpolicy==3.0.1`, and review the `.clawpolicy/` files it creates before allowing them to guide agent behavior.

A.I.G

Vulnerability Patterns

• Insecure Dependencies: Introduces malicious components through unsafe dependency sources
• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally

Findings (1)

T08 · Insecure Dependencies

Warning

• Location: scripts/verify_install.sh:8

Finding

Unpinned Third-Party Package Installation

Content

View full analysis

Vulnerability Details

File Location: scripts/verify_install.sh:8-10
Additional Locations: SKILL.md:35, SKILL.md:41, SKILL.md:60; references/upstream-README.md:20-34; references/upstream-README.zh-CN.md:20-34
Vulnerability Type: Unpinned and unaudited dependency installation
Risk Level: Medium

Vulnerable Code:

bash

source "$TMPDIR/venv/bin/activate"
python3 -m pip install --upgrade pip >/dev/null
python3 -m pip install clawpolicy >/dev/null

The documentation also recommends:

bash

python3 -m pip install clawpolicy
python3 -m pip install "clawpolicy[phase3]"

Technical Analysis

The verification script and installation documentation install clawpolicy without a version constraint or package hashes. Consequently, pip resolves whichever release is current at execution time rather than the release represented by the reviewed wrapper. The wrapper identifies itself as targeting version 3.0.1, but the installation command does not enforce that version.

Python package installation can execute package build logic, and subsequent smoke-test commands execute the installed package directly. Its transitive dependencies are also resolved without integrity pinning. The optional phase3 extra expands this uncontrolled dependency set further.

The pip upgrade is likewise unpinned. Although installation occurs inside a temporary virtual environment, code executed during installation and smoke testing retains the invoking user's operating-system privileges and can access resources available to that user. The external package implementation is not included in this project, so its behavior could not be verified by this audit.

Attack Path

1. An attacker compromises the clawpolicy PyPI project, one of its transitive dependencies, or a future package release.
2. The attacker publishes a malicious release that still satisfies the unconstrained package name.
3. A user or ...[truncated 1243 chars]

Remediation

View remediation

Remediation Suggestions

1. Pin the reviewed package release explicitly:

   bash

   python3 -m pip install "clawpolicy==3.0.1"
   

2. Generate a reviewed lock file containing exact versions and SHA-256 hashes for every transitive dependency, then enforce it:

   bash

   python3 -m pip install --require-hashes -r requirements.lock
   

3. Pin and hash optional dependencies separately. Do not recommend clawpolicy[phase3] until its expanded dependency graph has been audited and locked.

4. Remove the unconditional unpinned pip upgrade, or pin and verify the required pip version in the same manner.

5. Ensure the version declared in the wrapper metadata, documentation, verification script, and lock file remains synchronized.

6. Run package verification in a sandbox or isolated container with no credentials, minimal filesystem access, restricted outbound networking, and an unprivileged user. A virtual environment alone is not a security boundary.

7. Add automated dependency review, vulnerability scanning, provenance verification, and lock-file integrity checks to the release workflow.

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (1)

Natural-Language Policy Violations

Low

• Category: Not specified by scanner
• Confidence: 93% confidence

Finding

The document presents the current file entirely in Simplified Chinese and labels English as an alternate primary document, which is a natural-language locale choice embedded in the skill documentation. There is no user opt-in within this file and no stated region-specific reason for requiring or preferring this locale.

Content

No source excerpt is available for this finding.