Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

AgentShield

A scanner that grades a Claude Code setup for leaked keys, over-broad permissions, risky hooks and unsafe MCP servers.

Reviewed by ReviewRamble editors · Updated

View on GitHub Open source

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

3.8/5

AgentShield is the quickest way to find out how much your Claude Code setup trusts by accident, and its findings are a checklist to work through rather than a verdict to frame.

What is AgentShield?

AgentShield reads the configuration of a coding agent, mainly the ~/.claude directory and project .claude folders, and reports what could go wrong: API keys pasted into CLAUDE.md, allow rules such as Bash(*), hooks that interpolate file names into shell commands, MCP servers launched with npx -y, and instructions that tell the agent to fetch and run remote code. It prints an A to F grade with a 0 to 100 score and can fix the safest problems itself.

It runs with npx ecc-agentshield scan, as a GitHub Action that writes SARIF for code scanning, or inside the ECC Tools GitHub App. An optional --opus mode sends the configuration to Claude for a three-agent red team, blue team and auditor review. Affaan Mustafa built it at the Cerebral Valley and Anthropic Claude Code hackathon in February 2026 as part of his Everything Claude Code project.

The CLI is MIT-licensed and free; ECC Tools Pro, a GitHub App for organisation-wide analysis, costs $19 per seat a month.

Is AgentShield worth it?

AgentShield earns its place because it reads the files that matter in a coding agent setup, which general secret scanners do not understand. Its 268 rules across 15 modules cover permission allow lists, 40 hook patterns from reverse shells to clipboard exfiltration, 49 MCP server checks and 41 rules for agent and CLAUDE.md instructions. The score starts at 100 per category and only findings deduct from it, 25 for a critical and 15 for a high, so adding decorative deny rules cannot raise a grade.

Adoption is real if modest: about 1,413 GitHub stars, 66,299 npm downloads of ecc-agentshield in the 30 days to 15 November 2026, and v1.6.0 released on 10 September 2026. The GitHub Action, SARIF output, baselines that flag only new findings, and an evidence-pack export make it usable in CI rather than just on a laptop.

The noise is the main cost. The maintainer keeps a false-positive audit file in the repository, MCP findings in template folders are down-weighted to 0.25x, and the documentation spends several paragraphs explaining which findings are active runtime risk and which are examples. Expect to triage. Claude Code is the primary target; other harnesses such as Codex, OpenCode and Zed get evidence notes rather than the same depth.

The --opus mode sends your configuration to Anthropic, or to OrcaRouter with --provider orcarouter, and the README warns against using the latter on configs with unredacted secrets. Run the local scan first and fix what it finds before sending anything out. Compared with Snyk's agent scanner or NVIDIA's SkillSpector, which focus on skills and MCP tool poisoning, AgentShield is broader about the harness itself and less mature in its precision.

Where it earns its place

  • A solo developer with a year of accumulated Claude Code hooks and MCP servers

    Find out whether any of my config leaks keys or lets the agent run anything

    Run npx ecc-agentshield scan with no install, read the critical and high findings first, and apply --fix for hardcoded secrets. It costs nothing and finishes locally without sending files anywhere.

  • A platform team rolling out a shared .claude directory to 40 engineers

    Stop risky permissions or hooks being merged into the team config

    Add the affaan-m/agentshield@v1 GitHub Action with a baseline so pull requests fail only on new critical or high findings, and upload SARIF to code scanning. The Action is free; ECC Tools Pro at $19/seat/month adds organisation-wide analysis.

  • A security engineer preparing evidence for a SOC 2 review

    Show auditors how agent configurations are controlled

    Run the scan with --compliance and --evidence-pack to map findings to control IDs and export a portable bundle, then attach the HTML report to the review.

Before you commit

Setup
npx ecc-agentshield scan auto-discovers ~/.claude and prints a graded report in one command; agentshield init writes a stricter baseline config.
Integrations
GitHub Action with SARIF and job summaries; JSON, Markdown and HTML output; ECC Tools GitHub App; external rule packs; Anthropic API or OrcaRouter for --opus; npm registry lookups with --supply-chain-online.
Data and privacy
The default scan runs locally and supply-chain checks run offline. --opus sends configuration contents to Anthropic, or to OrcaRouter when selected, so redact secrets first.
Cost at scale
Free for the CLI and Action. --opus uses Claude Opus at your Anthropic API rates; ECC Tools Pro is $19 per seat a month.
Limits
False positives are common enough that the maintainer documents a triage process; template MCP findings are down-weighted to 0.25x; coverage is deepest for Claude Code.

What’s good

  • Understands Claude Code settings, hooks and MCP config
  • Score cannot be gamed by adding defences
  • GitHub Action with SARIF and baseline drift
  • Auto-fix moves hardcoded secrets to environment variables

What’s not

  • Many findings need manual triage
  • Depth drops outside Claude Code
  • --opus mode sends your config to a third-party API
  • No release since 10 September 2026

Best for

Developers and platform teams who run Claude Code with hooks and MCP servers and want a repeatable audit of that configuration in CI.

AgentShield alternatives

AgentShield: common questions

Is AgentShield worth it?
AgentShield is the quickest way to find out how much your Claude Code setup trusts by accident, and its findings are a checklist to work through rather than a verdict to frame. We rate it 3.8 out of 5.
How much does AgentShield cost?
CLI and GitHub Action free under MIT. The --opus analysis bills your Anthropic API key. ECC Tools Pro GitHub App: $19/seat/month.
What are the downsides of AgentShield?
Many findings need manual triage; depth drops outside Claude Code; --opus mode sends your config to a third-party API; no release since 10 September 2026.
Who is AgentShield best for?
Developers and platform teams who run Claude Code with hooks and MCP servers and want a repeatable audit of that configuration in CI.
How much does AgentShield cost for a team or in production?
Free for the CLI and Action. --opus uses Claude Opus at your Anthropic API rates; ECC Tools Pro is $19 per seat a month.
Is AgentShield safe for company data?
The default scan runs locally and supply-chain checks run offline. --opus sends configuration contents to Anthropic, or to OrcaRouter when selected, so redact secrets first.
What are the best AgentShield alternatives?
The closest alternatives we have reviewed are Snyk Agent Scan, SkillSpector, Cisco Skill Scanner and Everything Claude Code (ECC).

Answers checked Nov 16, 2026.