AgentIndex icon
AgentIndex
ToolsMCP ServersHOTCategoriesTrendingNewBest ListsCompareGuides
Submit Tool
ToolsMCP ServersHOTCategoriesTrendingNewBest ListsCompareGuides
Home/
Security & Safety/
SkillSpector
SkillSpector logo

SkillSpector

Active·★ 26.6k·Apache-2.0·Updated 2026-11-16
★ Most Popular

SkillSpector inspects AI agent skill bundles for prompt injection, credential access and other security risks before installation. It is relevant to teams reviewing external resources, especially when they need findings tied to files and source locations. We tested package 2.12.0 at a fixed commit in an isolated Python 3.12 environment. This is a commit-specific static-mode review, not a claim about the latest release, every scanner rule or optional model-assisted analysis. The package uses Apache-2.0 licensing and retains normal workflow dependencies even when model analysis is disabled.

Three handwritten skill files described a benign local task, deliberately unsafe instructions and a warning that explicitly prohibited the unsafe behavior. The benign example produced no issues and score zero. Both other examples triggered P1 instruction override and PE3 credential access, with score 42, overall MEDIUM severity and highest individual severity HIGH. With --fail-on-findings, both scans exited one. JSON retained finding evidence, and SARIF 2.1.0 contained two results. The warning example demonstrates why a static match should lead to contextual review rather than automatic classification of intent.

After reviewing the warning, we generated an explicit baseline with an explanatory reason and rescanned the same directory. Two findings were suppressed, no active issues remained and the gate returned zero. Suppression records acceptance; it does not repair the source or prove safety. All scans used --no-llm, and no unsafe specimen was executed or installed. We did not test dependency manifests, remote ingestion, CVE lookups, runtime containment or real-host behavior. Adopt it as one review step when findings and coverage limits will be examined, not as an unconditional installation approval.

Read the full review, verification scope and sources →

Hands-on tested · Tested on 2026-11-05 with 2.12.0; commit 62c2fe8622f580ac292d4078c3b71f790194dbbd (static mode only).

Setup: macOS arm64; Node.js 26.6.0, Chrome 156.0.8078.82, CPython 3.12.13; isolated checkouts and synthetic fixtures

Commands run

python3 doc/review-fixtures/skill-selection-20261105/verify.py --sources <source-parent> --python <venv>/bin/python --output /tmp/skill-review-results.json

What we saw

  • Benign sample: zero findings; unsafe and prohibited-behavior warning: P1/PE3, score 42, exit 1 with the findings gate.
  • SARIF 2.1.0 contained two results; reviewed warning baseline suppressed two findings with zero active issues.
  • Static --no-llm only. No skill execution, model, remote input, dependency or CVE test.

Data as of 2026-11-12.

License
Apache-2.0
Latest release
v2.13.4 · 2026-11-04
Last commit
2026-11-11
Contributors
125
Forks
2,245
Open issues and PRs
151
First published
2026-03-21

Source: GitHub, npm and PyPI public APIs.

#agent-security#agent-skills#agentic-ai#ai-security#claude-code#dev-tooling#mcp#prompt-injection
$ Install
$ uv tool install git+https://github.com/NVIDIA/skillspector.git
↗ Visit site★ GitHub
01

Features

01Static analysis with optional LLM semantic checks
02Repository, URL, archive and local-file inputs
03JSON, Markdown and SARIF reporting
04Baselines and suppression for accepted findings
02

Why choose it

+Static mode located instruction-override and credential-access patterns in the unsafe specimen.
+JSON evidence, SARIF 2.1.0 output and a findings-based exit gate were exercised successfully.
+An explicit reviewed baseline suppressed two warning findings and retained the acceptance decision.
03

Trade-offs

!Quoted prohibited behavior triggered the same static findings and score as unsafe instructions.
!A zero score or accepted baseline is not proof of safety or runtime containment.
!Model-assisted accuracy, remote ingestion, dependencies and CVE lookup were not tested.
05

Maintenance check

ActiveChecked: 2026-11-05
Last commit
2026-11-05
Commits, last 90 days
100+
Releases, last 90 days
18
Latest release
v2.13.4 (2026-11-04)
Open issues
178
Recent issues with at least one comment
1/13
Commits per week, last 12 weeks
Commits per week, last 12 weeks: 26, 26, 19, 24, 49, 58, 103, 176, 43, 11, 53, 119

Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.

Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.

Most discussed open issues

  • #596 AE1 classifies opaque-by-type assets as `analysis-evasion` HIGH, discarding the ledger `reason_code` that distinguishes them from real cover8 comments · 2026-09-20

Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.

04

Who it fits

Pick it if
  • You need an additional pre-installation review step with source locations.
  • You can review findings and document narrow baseline decisions.
Look elsewhere if
  • You need automatic proof that a skill is safe to execute.
  • You expect static matching to reliably distinguish quotation and intent.
06

Compatibility

Python
3.12+
Verified via docs
Docker
Documented container build and scan workflow
Verified via docs
07

Quick start

1
$ uv tool install git+https://github.com/NVIDIA/skillspector.git
08

Use cases

↳Reviewing third-party skills before installation
↳Adding skill scans to CI workflows
↳Tracking new findings against a baseline
09

Alternatives

⚡Top VS:
VSrepo-forensics★208→VSskills★4.0k→VStaste-skill★117.7k→VSSkillEvaluator★3.3k→
repo-forensics logo
repo-forensics★ 208
Offline security scanner for AI-agent repos, skills, plugins, and MCP servers.
VSCompare→
skills logo
skills★ 4.0k
Agent Skills for NVIDIA products — install into Claude Code, Codex, and other coding agents to run Physical AI, robotics, simulation, CUDA, and RAG workflows end to end.
VSCompare→
taste-skill logo
taste-skill★ 117.7k
Agent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages
VSCompare→
SkillEvaluator logo
SkillEvaluator★ 3.3k
Multi-tier framework for evaluating AI agent skills with quality gates, semantic overlap detection, synthetic evaluation dataset generation, and live agent evaluation that measures how skills affect agent behavior.
VSCompare→
diagram-design logo
diagram-design★ 67.9k
Editorial diagram design for Claude Code, Codex, GitHub Copilot, Cursor, Factory Droid, Qoder, and Pi. 45 diagram types. Self-contained HTML + SVG. No shadows. No Mermaid slop.
VSCompare→
skills logo
skills★ 57.8k
Skills for Designers and Engineers.
VSCompare→
McpVanguard logo
McpVanguard★ 14
An open-source security proxy and active firewall for the Model Context Protocol (MCP). It acts as a real-time 'Reflex System' between AI agents and their tools, protecting the host system from malicious intent, prompt injection, and data exfiltration.
brag logo
brag★ 21.8k
You built it. Now brag. Turn the project you just created into a short, shareable launch video with one command.
VSCompare→
See all alternatives →

Comments

Log in to leave a comment

No comments yet. Be the first!

On this page
01Features02Why choose it03Trade-offs05Maintenance check04Who it fits06Compatibility07Quick start08Use cases09Alternatives
NVIDIA /
SkillSpector
↗

Pre-installation skill triage that still needs contextual review

26.6k2.2kApache-2.0
Last commit2d ago
StatusActive
CategorySecurity & Safety
Contributors125
Latest releasev2.13.4
Links
Documentation↗Discussion↗Issues↗Releases↗
Cloud Infrastructure
Ad

Deploy on DigitalOcean

Host your MCP servers, Docker containers, and AI agents with $200 free credit over 60 days.

Claim $200 Free Credit
© 2026 AgentIndex.app|Built by a 10-year iOS Developer.
Best by use caseAboutContactAdvertisePrivacy PolicyTerms of UseQYSGitHubBuy me a coffee ☕

Browse by Category

MCP ServersCode AssistantWorkflow AutomationRAG / Knowledge BaseMulti-AgentBrowser AutomationLLM InfraDev ToolingObservability

Not affiliated with Anthropic, OpenAI or Microsoft.