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/rfunix/tengu/claude-mdgit clone --depth 1 https://github.com/rfunix/tenguWrote 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/rfunix/tengu/claude-md)<a href="https://agentmods.dev/instructions/rfunix/tengu/claude-md"><img src="https://agentmods.dev/badge/instructions/rfunix/tengu/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 | $0.06055 | $0.06055 |
| Opus 5 | $0.03027 | $0.03027 |
| Sonnet 5 | $0.01211 | $0.01211 |
| Haiku 4.5 | $0.00605 | $0.00605 |
Grade A, and why
tengu CLAUDE.md scanned grade A with 2 findings 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 5d 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.
| curl (internal) | `-x` | Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
- Use `asyncio.create_subprocess_exec` (via `run_command`), never `subprocess.run` or `shell=True`. How it starts
The opening of the file, as written. The whole thing — 662 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md — Tengu MCP Server
This file is the primary guide for AI assistants (Claude Code and others) working on the Tengu codebase. Read it in full before making any changes.
Project Overview
Tengu is a Model Context Protocol (MCP) server that exposes professional pentesting tools to AI assistants through a clean, secure interface.
| Property | Value |
|---|---|
| Framework | FastMCP 2.0+ |
| Python | 3.12+ |
| Package manager | uv (full path: /Users/rfunix/.local/bin/uv) |
| Validation | Pydantic v2 |
| Logging | structlog (JSON, structured) |
| Entry point | src/tengu/server.py → FastMCP("Tengu") |
| Config file | tengu.toml at project root |
| Test suite | 2643+ tests, 0 lint errors |
| Tools | 80 MCP tools |
| Resources | 20 MCP resources |
| Prompts | 35 MCP prompts |
Architecture
Request Flow
Every tool invocation follows this mandatory pipeline — no exceptions:
Claude (MCP Client)
│
▼
FastMCP layer ← receives JSON-RPC call
│
▼
1. sanitizer ← tengu/security/sanitizer.py
- Validate input formats (IP, hostname, CIDR, URL, port spec, etc.)
- Strip/reject shell metacharacters: [;&|`$<>(){}\[\]!\\\'\"\\r\\n]
- Raise InvalidInputError on any violation
│
▼
2. allowlist ← tengu/security/allowlist.py
- Extract hostname from target
- Check against blocked_hosts (always wins)
- Check against allowed_hosts (must match if non-empty)
- Raise TargetNotAllowedError on denial
│
▼
3. rate_limiter ← tengu/security/rate_limiter.py
- Sliding window (60s) per tool
- Concurrent slot limit per tool
- Raise RateLimitError if either limit is exceeded
│
▼
4. audit logger ← tengu/security/audit.py
- Write JSON audit record to logs/tengu-audit.log
- Redact sensitive parameters (password, token, key, etc.)
- Log tool name, target, params, result, duration
│
▼
5. executor ← tengu/executor/process.py
- asyncio.create_subprocess_exec (NEVER shell=True)
- Resolve tool path via shutil.which
- Enforce timeout
- Return (stdout, stderr, returncode)
│
▼
Output parser ← per-tool parsing function
(XML, JSON, JSONL, plain text)
│
▼
Return dict to Claude
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.
- 5d ago First seen · 662 lines · 6,055 tokens per session scan A 23dfade771d9
tengu CLAUDE.md is an instructions file published in the GitHub repository rfunix/tengu (56 stars, last pushed 2mo ago), licensed MIT. It adds 6,055 tokens to every session, about $0.0303 per session on Opus 5. A static security scan graded it A with 2 findings (makes network calls, runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
toolhive-studio copilot-instructions.md
Copilot instructions for stacklok/toolhive-studio, covering instructions for copilot, ui components, data fetching, file naming conventions and testing.
toolhive-studio CLAUDE.md
Claude Code instructions for stacklok/toolhive-studio, covering project overview, build and development commands, environment requirements, common gotchas and auto-generated code — never edit manually.
toolhive-studio AGENTS.md
AGENTS.md instructions for stacklok/toolhive-studio, covering project overview, essential commands, common gotchas, boundaries and code conventions.
agent-security-scanner-mcp CLAUDE.md
Claude Code instructions for sinewaveai/agent-security-scanner-mcp, covering claude.md, build and development commands, install dependencies, run all tests (28 test files, 420+ tests, 4.5 min) and run specific test file.
assay CLAUDE.md
Claude Code instructions for Rul1an/assay, covering assay - ai agent context, what is assay?, workspace structure, key commands and cli entry points.
assay AGENTS.md
AGENTS.md instructions for Rul1an/assay, covering assay agent contract, canonical state, adr-042/043 scope, branch and worktree ownership and development discipline.