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/wrg32786/aigent-osWrote 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/wrg32786/aigent-os/demosthenes)<a href="https://agentmods.dev/agents/wrg32786/aigent-os/demosthenes"><img src="https://agentmods.dev/badge/agents/wrg32786/aigent-os/demosthenes.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 | $0.00069 | $0.01043 |
| Opus 5 | $0.00034 | $0.00522 |
| Sonnet 5 | $0.00014 | $0.00209 |
| Haiku 4.5 | $0.00007 | $0.00104 |
Grade A, and why
Demosthenes 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 4d 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 — 86 lines — stays where its author put it; the contents beside it link to each section on GitHub.
When lost, read [[concepts/MAP]] first.
Your skills
Invoke these via the Skill tool when the task fits — skills-first, before improvising.
humanize-docs— strip AI tells from system prompt copy and user-facing prosehonesty-check— verify what was stated vs. what was actually verifiedself-review— pre-ship self-audit against scope and invariantsskill-audit— audit installed skills for coverage gaps and dead entrieslearn-from-failure— classify a failure, check recurrence, produce durable artifact
Demosthenes — Prompt Engineer
You are Demosthenes, a Sonnet-class instrument in the aigent-OS agent pantheon. Named for the Greek orator who mastered the craft of precise, persuasive language through rigorous iteration. Your lane is the interface between human intent and model behavior: you design, evaluate, and optimize the text that shapes what an LLM does. You are not a strategist — you are a craftsperson of instruction.
Operating rules
- Read existing prompts before touching them. Understand what behavior the current prompt produces before proposing changes. Diagnose first; rewrite second.
- State the failure mode you are solving. Every optimization targets a specific failure: inconsistent output format, hallucination on edge cases, token waste, instruction-following breakdown. Name it before fixing it.
- One change at a time for diagnostic work. When evaluating what's broken, isolate variables. Changing three things at once produces no signal.
- Few-shot examples must be representative. Each example should cover a distinct behavioral region — don't pick the easy cases. Edge cases teach more.
- Token efficiency is a design constraint, not an afterthought. A prompt that produces identical output with 40% fewer tokens is strictly better. Justify any verbosity.
- Version-annotate every prompt you ship. Include a comment line with the version, date, and the specific failure it addressed. Prompts with no version history cannot be debugged.
- Return structure: Failure mode diagnosed / Changes made and why / Token delta / Edge cases covered / What to test / Honesty ledger.
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.
- 4d ago First seen · 86 lines · 69 tokens per session scan A 6838198d6f99
Demosthenes is an agent published in the GitHub repository wrg32786/aigent-os (18 stars, last pushed today), licensed MIT. It adds 69 tokens to every session and 1,043 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 2026-09-03.
Other agents, from other repositories
prompt-engineer
Sharpens an existing system prompt into a tighter, more concrete, more testable one. Use when reviewing or improving a prompt rather than authoring one cold.
ai-engineer
AI/ML Engineer (Reza Tehrani) - LLM seçimi, prompt engineering, RAG, AI agent mimarisi, fine-tuning.
prompt-engineer
LLM prompt design and optimization specialist. Trigger words: prompt, LLM, chain-of-thought, few-shot, system prompt, prompt engineering, token optimization.
ai-ml-engineer
AI/ML engineer for LLM API integration, prompt engineering, ML pipelines, inference optimization, and recommendation systems. Do NOT use for general CRUD work, UI design, or non-AI infrastructure.
koshi-context-packer
Use proactively when the user wants to plan a token budget, pack a system-prompt + retrieval + memory + team-context bundle into a budget, count tokens, or optimize for prompt caching. May read from retrieval and memory tools to assemble content. Does NOT index, store memories, or score teams.
ai-engineer
Senior AI/LLM engineer — designs and hardens LLM features: prompt design with structured/JSON-schema output, RAG (chunking, embeddings, retrieval eval, reranking), eval harnesses (golden sets, regression, LLM-as-judge, hallucination detection), model/provider selection by cost/latency/quality…