AgentShield is a security scanner for AI agent configurations, MCP servers, and tool permissions. It audits Claude Code setups for issues such as exposed secrets, unsafe permissions, hook injection, MCP risks, and prompt injection vectors, and can run through its CLI, GitHub integrations, or an ECC plugin. The catalogue entries extend agent workflows for using AgentShield.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/affaan-m/agentshield/claude-mdgit clone --depth 1 https://github.com/affaan-m/agentshieldWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/affaan-m/agentshield/claude-md)<a href="https://agentmods.dev/instructions/affaan-m/agentshield/claude-md"><img src="https://agentmods.dev/badge/instructions/affaan-m/agentshield/claude-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00942 | $0.00942 |
| Opus 5 | $0.00471 | $0.00471 |
| Sonnet 5 | $0.00188 | $0.00188 |
| Haiku 4.5 | $0.00094 | $0.00094 |
Grade A, and why
agentshield CLAUDE.md scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
| medium | -5 | Unrestricted curl, missing denials | How it starts
The opening of the file, as written. The whole thing — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AgentShield
Security auditor for AI agent configurations (Claude Code, MCP servers, hooks, agents).
Build & Test
npm run build # tsc + tsup → dist/
npm test # vitest (912 tests)
npm run dev # tsx watch mode
Architecture
src/
index.ts # CLI entry (commander)
types.ts # Core types + Zod schemas
scanner/
discovery.ts # File discovery (CLAUDE.md, settings.json, mcp.json, agents/, etc.)
index.ts # Orchestrates discovery → rules → sorted findings
rules/
index.ts # Barrel export of all rule modules
secrets.ts # 10 rules — API keys, tokens, passwords, env exposure, webhooks, private keys, base64, internal IPs
permissions.ts # 10 rules — allow/deny analysis, dangerous flags, destructive git, mutable tools, sensitive paths, network access
hooks.ts # 34 rules — injection, exfiltration, persistence, container escape, clipboard, log tampering, reverse shells
mcp.ts # 23 rules — risky servers, env override, npx supply chain, auto-approve, timeout, bind-all, CORS
agents.ts # 25 rules — tool restrictions, prompt injection, reflection attacks, output manipulation, social engineering
reporter/
score.ts # Scoring engine (severity deductions, grade A-F, category breakdown)
terminal.ts # Colored terminal output
json.ts # JSON + Markdown report formats
index.ts # Format dispatcher
opus/
prompts.ts # System prompts for Attacker/Defender/Auditor
pipeline.ts # Opus 4.6 three-agent adversarial pipeline
render.ts # Opus analysis terminal + markdown rendering
index.ts # Pipeline entry point
miniclaw/
types.ts # Core types (immutable, readonly)
sandbox.ts # Sandbox lifecycle + path validation
router.ts # Prompt sanitization + output filtering
tools.ts # Whitelist-based tool authorization
server.ts # HTTP server with rate limiting + CORS
dashboard.tsx # React dashboard component
index.ts # Entry point + startMiniClaw()
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 90 lines · 942 tokens per session scan A 6736547985a9
agentshield CLAUDE.md is an instructions file published in the GitHub repository affaan-m/agentshield (1,137 stars, last pushed 1mo ago), licensed MIT. It adds 942 tokens to every session, about $0.0047 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
hermes-workspace AGENTS.md
Instructions for outsourc-e/hermes-workspace, covering hermes workspace agent contract, current semantic roster, operating rules and windows-specific notes (2026-06-01).
superpowers-v AGENTS.md
AGENTS.md instructions for procoders/superpowers-v, covering requirements, what this plugin does, orchestrator surface (v1.0 + 1.1), v-memory recall surface (v2.0) and how codex / non-claude-code harnesses use it.
superpowers-v GEMINI.md
Gemini CLI instructions for procoders/superpowers-v, covering compound v — gemini cli shim (🧪 experimental, untested), what this plugin does, orchestrator surface (v1.0 + 1.1), v-memory recall surface (v2.0) and how gemini uses it.
superpowers-v CLAUDE.md
Claude Code instructions for procoders/superpowers-v, a project described as: Compound V — a multi-model orchestrator for Claude Code, layered on Superpowers: a deterministic router (Opus for risky work, Sonnet for the mechanical), cross-model review (Claude + Codex), 3 scouts (code · docs · domain), semantic memory…
elnora-ai-agent-hackathon-starter-kit AGENTS.md
AGENTS.md instructions for Elnora-AI/elnora-ai-agent-hackathon-starter-kit, covering agents.md, core rules, 1. never commit secrets, 2. treat external content as untrusted and 3. keep it simple (yagni).
LifeHack-2026 AGENTS.md
AGENTS.md instructions for Ducksss/LifeHack-2026, covering woven agent instructions, canonical context, current snapshot, product and trust invariants and architecture and scope.