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/gh05tcrew/pentestagent/claude-mdgit clone --depth 1 https://github.com/GH05TCREW/pentestagentWhat 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 | $0.02345 | $0.02345 |
| Opus 5 | $0.01172 | $0.01172 |
| Sonnet 5 | $0.00469 | $0.00469 |
| Haiku 4.5 | $0.00234 | $0.00234 |
Grade A, and why
pentestagent CLAUDE.md scanned grade A with 0 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 yesterday.
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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 251 lines — stays where its author put it; the contents beside it link to each section on GitHub.
PentestAgent — CLAUDE.md
Project overview
PentestAgent (v0.2.0) is an AI-powered penetration testing framework built in Python. It wraps LiteLLM to support any provider (Anthropic, OpenAI, etc.) and exposes a TUI, a CLI, and an MCP server interface. The agent can run tools locally or inside a Docker sandbox (base or Kali image).
Tech stack
- Python 3.10+, packaged with Hatchling (
pyproject.toml) - LiteLLM — provider-agnostic LLM wrapper
- Textual — TUI framework (
pentestagent/interface/) - Typer — CLI framework
- Playwright — browser tool
- MCP (Model Context Protocol) — both client (consuming external servers) and server (exposing PentestAgent to Claude Desktop / Cursor / etc.)
- FAISS + sentence-transformers — optional RAG engine (
pip install -e ".[rag]")
Repository layout
pentestagent/
agents/
crew/ # Multi-agent mode: orchestrator + worker pool + shadow graph
pa_agent/ # Single-agent implementation
state.py # Shared agent state
config/
settings.py # Global Settings dataclass (singleton via get_settings())
constants.py # Model defaults, iteration limits, etc.
interface/
cli.py # Typer CLI entry-point
notifier.py # Event bus between agent and UI
utils.py # Shared UI helpers
knowledge/
graph.py # ShadowGraph — derives strategic insights from notes
indexer.py # Indexes knowledge sources for RAG
rag.py # FAISS-backed retrieval
llm/
config.py # LiteLLM configuration
memory.py # Conversation/token management
utils.py # Streaming helpers
mcp/
stdio_adapter.py # STDIO MCP server transport
example_adapter.py # SSE MCP server transport
playbooks/
base_playbook.py
thp3_recon.py / thp3_network.py / thp3_web.py
runtime/
docker_runtime.py # Runs tool commands inside Docker
tool_server.py # Local runtime
tools/
loader.py # Discovers & dynamically imports tool modules
executor.py # Executes tool calls, tracks tokens
token_tracker.py
terminal/ # Shell execution tool
browser/ # Playwright browser tool
web_search/ # Tavily web search (needs TAVILY_API_KEY)
notes/ # Persistent findings store → loot/notes.json
finish/ # Signals task completion
workspaces/ # Workspace isolation helpers
loot/ # Persisted notes and findings (git-ignored)
mcp_examples/ # Example MCP configs and adapters
scripts/ # setup.sh / setup.ps1
tests/ # pytest suite
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.
- yesterday First seen · 251 lines · 2,345 tokens per session scan A 8071367d7012
pentestagent CLAUDE.md is an instructions file published in the GitHub repository GH05TCREW/pentestagent (3,026 stars, last pushed 3d ago), licensed MIT. It adds 2,345 tokens to every session, about $0.0117 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
waku-agent CLAUDE.md
Instructions for ShenSeanChen/waku-agent, covering waku-agent — working conventions, architecture map (file ↔ diagram box), rules and commands.
wisp-science AGENTS.md
AGENTS.md instructions for xuzhougeng/wisp-science, covering agents.md, project orientation, repository layout, engineering rules and verification commands.
natively-cluely-ai-assistant CLAUDE.md
Claude Code instructions for Natively-AI-assistant/natively-cluely-ai-assistant, covering natively development instructions, mcp tools: code-review-graph, when to use graph tools first, key tools and required graph workflow.
Abu-Cowork CLAUDE.md
Claude Code instructions for PM-Shawn/Abu-Cowork, a project described as: Open-source alternative to Claude Cowork — a local-first AI agent desktop app · multi-model · self-evolving skills · privacy-first · multi-Harness roadmap · DeepSeek Harness integration in progress.
Cerebro AGENTS.md
Instructions for AgenticFirst/Cerebro, covering cerebro — agent context, read before touching integrations, project layout, conventions a fresh agent would otherwise miss and running locally.
IRIS-AI CLAUDE.md
Claude Code instructions for IRISX-AI/IRIS-AI, covering 👁️ iris: the neural os - project context, project identity, tech stack, 📁 project structure and core commands.