Trust Assessment
model-pruning received a trust score of 65/100, placing it in the Caution category. This skill has some security considerations that users should review before deployment.
SkillShield's automated analysis identified 5 findings: 0 critical, 1 high, 2 medium, and 2 low severity. Key findings include Network egress to untrusted endpoints, Covert behavior / concealment directives, Unpinned `git clone` from external repositories.
The analysis covered 4 layers: Manifest Analysis, Static Code Analysis, Dependency Graph, LLM Behavioral Safety. All layers scored 70 or above, reflecting consistent security practices.
Last analyzed on February 12, 2026 (commit 458b1186). SkillShield performs automated 4-layer security analysis on AI skills and MCP servers.
Layer Breakdown
Behavioral Risk Signals
Security Findings5
| Severity | Finding | Layer | Location | |
|---|---|---|---|---|
| HIGH | Unpinned `git clone` from external repositories The installation instructions use `git clone` to fetch `wanda` and `sparsegpt` repositories directly from GitHub without specifying a commit hash or tag. This means the user will always get the latest `main` branch, which can change over time. A malicious actor gaining control of these repositories could introduce harmful code that would then be installed and executed by users of this skill. The `pip install -r requirements.txt` and `pip install -e .` commands further exacerbate this by executing code from the unverified cloned content. Pin specific commit hashes or tags for `git clone` operations (e.g., `git clone -b <tag_or_commit_hash> https://github.com/locuslab/wanda`). Alternatively, consider using official PyPI packages if available, or vendoring the code with proper review. | Static | SKILL.md:24 | |
| MEDIUM | Network egress to untrusted endpoints HTTP request to raw IP address Review all outbound network calls. Remove connections to webhook collectors, paste sites, and raw IP addresses. Legitimate API calls should use well-known service domains. | Manifest | cli-tool/components/mcps/devtools/figma-dev-mode.json:4 | |
| MEDIUM | Unpinned dependencies in manifest and `pip install` commands The skill's manifest lists `transformers` and `torch` without specific version pinning. Similarly, the `pip install` commands for `torch`, `transformers`, and `accelerate` do not specify versions. This can lead to non-reproducible environments, unexpected breaking changes, or the inadvertent installation of versions with known vulnerabilities if a new version is released with a critical flaw. Pin specific versions for all dependencies in the manifest and `pip install` commands (e.g., `transformers==4.30.0`, `torch==2.0.0`). Use a `requirements.txt` file with pinned versions for all direct and transitive dependencies. | Static | SKILL.md:2 | |
| LOW | Covert behavior / concealment directives Multiple zero-width characters (stealth text) Remove hidden instructions, zero-width characters, and bidirectional overrides. Skill instructions should be fully visible and transparent to users. | Manifest | cli-tool/components/mcps/devtools/jfrog.json:4 | |
| LOW | Missing dependency `lm_eval` The evaluation section of the skill imports `lm_eval` (`from lm_eval import evaluator`) but this package is not listed in the skill's manifest dependencies nor included in the installation instructions. Users attempting to run the evaluation code will encounter an `ImportError`. If they then try to install it manually, they might be susceptible to typosquatting attacks if they misspell the package name or if a malicious package with a similar name exists. Add `lm_eval` to the `dependencies` list in the skill's manifest and include it in the installation instructions, preferably with a pinned version. | Static | SKILL.md:293 |
Scan History
Embed Code
[](https://skillshield.io/report/27bbc107ed9fd306)
Powered by SkillShield