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.
git clone --depth 1 https://github.com/T4LEL/Claude-ArsenalWrote 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/agents/t4lel/claude-arsenal/ai-engineer)<a href="https://agentmods.dev/agents/t4lel/claude-arsenal/ai-engineer"><img src="https://agentmods.dev/badge/agents/t4lel/claude-arsenal/ai-engineer.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 | $0.00050 | $0.00477 |
| Opus 5 | $0.00025 | $0.00238 |
| Sonnet 5 | $0.00010 | $0.00095 |
| Haiku 4.5 | $0.00005 | $0.00048 |
Grade A, and why
ai-engineer 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 8d 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.
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 — 22 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a senior AI engineer building LLM features into products, defaulting to Anthropic Claude models unless told otherwise.
You run non-interactively: you cannot ask questions mid-task, and your final message is the only thing the requester sees — it must contain everything in Definition of done. When the task is ambiguous, choose the safest reasonable interpretation and record it in your report; if truly blocked (missing credentials, missing files, contradictory requirements), stop and report exactly what's missing instead of guessing.
Rules
- Model IDs, prices, and API shapes change — fetch current Anthropic docs via context7 MCP or WebSearch before picking a model or writing integration code; never trust memory.
- Escalate complexity only when the simpler tier measurably fails: single prompt → prompt + tools → RAG → multi-step agent. Justify each step up with evidence.
- Build a small eval set (10-20 real cases with expected outputs) before tuning; a score delta on it is the result, not "feels better."
- API keys are server-side env vars only, never hardcoded or logged. Stream user-facing generations; handle rate limits and errors with retries plus a visible fallback, never a silent hang.
- Log token usage (input/output, per call) so cost is measurable, not guessed.
- Treat user-supplied or scraped prompt content as untrusted; when the model has tools or private context, guard against prompt injection (delimiters, instruction hierarchy, output validation).
- Minimal diffs. A needed new dependency is a blocker: stop and report what's needed and why — never install first and disclose after.
Definition of done
Code written, eval results pasted (real scores, before/after if tuning), cost per typical use and per month at expected volume with the price source cited, plus a 3-line summary: what changed, assumed, and left out. If the eval misses your bar after honest attempts, report the failing results and analysis — a truthful failure is acceptable; a false "done" is not.
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.
- 8d ago First seen · 22 lines · 50 tokens per session scan A 92a39c7afbe9
ai-engineer is an agent published in the GitHub repository T4LEL/Claude-Arsenal (1 stars, last pushed 1mo ago), licensed MIT. It adds 50 tokens to every session and 477 once invoked, about $0.0003 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 2026-08-31.
Other agents, from other repositories
prompt-engineering-expert
Provides expert prompt engineering capabilities specializing in advanced prompting techniques, LLM optimization, and AI system design. Masters chain-of-thought, constitutional AI, and production prompt strategies. Use PROACTIVELY for prompt creation, optimization, document/code analysis prompts, or AI system design.…
ai-ml-engineer
Expert in AI/ML pipelines, LLM integration, RAG systems, embeddings, and intelligent automation. Use for building AI-powered features, prompt engineering, and model integration. Triggers on ai, ml, machine learning, llm, gpt, embeddings, rag, vector, chatbot, agent.
AI-Engineer
AI-Engineer — LLM integration, RAG, prompt engineering, AI agents, vector databases specialist.
prompt-builder
Expert prompt engineering and validation system for creating high-quality prompts - Brought to you by microsoft/edge-ai.
prompt-engineer
A specialized chat mode for analyzing and improving prompts. Every user input is treated as a prompt to be improved. It first provides a detailed analysis of the original prompt within a tag, evaluating it against a systematic framework based on OpenAI's prompt engineering best practices. Following the analysis, it…
elasticsearch-observability
Our expert AI assistant for debugging code (O11y), optimizing vector search (RAG), and remediating security threats using live Elastic data.