REPOSITORY EVIDENCE / INDEXABLE REPORT

addyosmani/agent-skills — MCP & Agent Security Evidence Report

A reproducible, bounded ShadowMCP review of public repository evidence. This page records what the selected scan observed, what it did not observe in scope, and what deserves human review before execution or approval.

Is addyosmani/agent-skills safe to run?

ShadowMCP does not issue a safe/unsafe verdict. This report shows bounded repository evidence that can help identify what deserves review before running or approving the project.

Scan identity

This report is tied to a specific repository revision and scanner version. It is a time-bounded evidence record, not a permanent statement about the repository.

Report schemashadowmcp.repo-evidence.v9
Scannershadowmcp.web-repo-scan.v22
Scan completed2026-09-17T19:43:27.163Z
Repository branchmain
Scanned commitbe4e44a9fbc5e8df0beaefadbb28bd22ee61cc39
Repository updated2026-09-17T19:36:06Z

What the bounded scan found

Review priority
Priority review

This is a triage priority, not a repository risk rating. Multiple evidence classes deserve contextual review.

No MCP configuration evidence was observed in the selected bounded repository scan. That does not prove MCP is absent elsewhere, including other branches, deeper paths, local client configuration or runtime environments.

The scan observed agent instruction/configuration signals, CI/CD automation and documentation-level download/install command markers. It did not observe shell/command-execution evidence in the inspected implementation set. README examples such as npx and git clone are documentation evidence, not proof that repository code executed those commands.

Evidence observed is not risk proven. Agent files, workflow files and command examples can be normal for the repository's purpose. ShadowMCP identifies review surfaces; it does not label them malicious.

Evidence categories

Shell / command execution

Not found in scanned scope

No matching shell/command-execution evidence was observed in the selected bounded implementation evidence.

Download / install

Found

Documentation command markers were observed; this does not mean ShadowMCP or repository code executed them.

  • README.mdnpx · git clone

MCP configuration

Not found in scanned scope

No matching MCP configuration evidence was observed in the selected bounded scan. This is not proof of absence outside that scope.

Agent permissions / behavior

Found

Agent-related instruction, model, tool/permission and plugin markers were observed in selected files.

  • AGENTS.mdagent instruction/config file · hooks · agent instructions · model config
  • CLAUDE.mdagent instruction/config file · tools / permissions · hooks · agent instructions
  • .claude-plugin/marketplace.jsonagent instructions
  • .claude-plugin/plugin.jsonagent instructions
  • scripts/run-evals-test.jsagent instructions
  • scripts/run-evals.jstools / permissions · agent instructions
  • scripts/validate-artifact-paths-test.jsagent instructions
  • scripts/validate-commands-test.jsagent instructions
  • scripts/validate-commands.jsagent instructions
  • scripts/validate-reference-links-test.jsagent instructions
  • scripts/validate-reference-links.jsagent instructions

CI/CD automation

Found

Automation is present and deserves contextual review of triggers, permissions, secrets, third-party actions and deployment behavior.

  • .github/workflows1 direct workflow file

Supply-chain / dependencies

Not found in scanned scope

No selected dependency evidence was observed in this bounded scan. This does not prove that the repository has no dependencies.

What to review first

  1. AGENTS.md and CLAUDE.md: confirm intended agent instructions, hooks, model configuration and requested tool/permission behavior.
  2. .claude-plugin/*: review plugin metadata and instructions before enabling the plugin in a compatible client.
  3. .github/workflows: inspect triggers, permissions, third-party actions, secret use and deployment or publishing steps.
  4. README.md command examples: understand what npx and clone/install instructions would fetch or run before following them.
  5. Selected scripts: inspect agent-related validation/evaluation scripts in context rather than treating markers as vulnerabilities.

Scan scope

The report used ShadowMCP's bounded root + selected depth-1 model. It did not perform a general recursive crawl.

Scope itemRecorded value
ModeBOUNDED_ROOT_AND_SELECTED_DEPTH1
Selected maximum depth1
Selected root files2
Selected depth-1 directories5
Selected depth-1 files10
Skipped selected depth-1 files0
Maximum readable file size250,000 bytes

State semantics: FOUND means selected evidence was observed; it does not prove a vulnerability, malicious behavior or unsafe operation. NOT FOUND IN SCANNED SCOPE means matching evidence was not observed by the bounded checks; it does not prove absence elsewhere. NOT CHECKED is used when an evidence class or identity could not be meaningfully checked.

Repository context

These public signals describe the repository at scan time. Popularity, ownership, activity and license metadata are not security guarantees.

Age / activitycreated 2026-02-15 · 213 days old · last push 2026-09-12 · 5 days since last push
Community signals95,747 stars · 10,143 forks · 126 open issues
OwnershipUser · addyosmani
LicenseMIT
Repository statearchived NO · fork NO · default branch main

Limitations

  • Public, selected and bounded repository evidence only.
  • FOUND means evidence observed, not risk or malicious behavior proven.
  • NOT FOUND IN SCANNED SCOPE is not proof of absence outside the bounded scan.
  • Trust/community metadata is descriptive context, not a security guarantee.
  • Zero selected dependency evidence does not mean zero dependencies.
  • No repository clone or target-code execution.
  • No raw source-file bodies or secret/token values are included.
  • No general recursive crawl.
  • No CVE lookup, package reputation lookup or package registry lookup.
  • No numeric security score and no safe/unsafe verdict.

Continue the review

This is one bounded evidence snapshot, not a permanent certificate. If the repository changes after the recorded commit, rescan it and compare the new evidence against this revision.

GUIDE

MCP Security Guide

Inventory, authorization, tool trust, supply chain, approval and drift.

CHECKLIST

MCP Server Security Checklist

Concrete server-side controls before approval or deployment.

DISCOVERY

Shadow MCP Discovery

Approved inventory, bounded evidence and runtime visibility are different things.

TOOL

Scan a public repository

Generate current evidence for another public GitHub repository.