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/inarus/provgraf/agents-mdgit clone --depth 1 https://github.com/Inarus/provgrafWhat 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 | $0.00887 | $0.00887 |
| Opus 5 | $0.00443 | $0.00443 |
| Sonnet 5 | $0.00177 | $0.00177 |
| Haiku 4.5 | $0.00089 | $0.00089 |
Grade A, and why
provgraf AGENTS.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 — 69 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Working with this bank as an agent
This repository is a bank of verified facts. The point of it is that a number you find here can be traced to the document or the person it came from, and that nothing enters it silently. That constraint applies to you.
The one rule
Nothing enters the bank without provenance and without a human's approval. You may read, search, propose and prepare a change. You do not decide what is true. If you are unsure whether something qualifies as certain, it does not — say so rather than writing it.
Reading
The MCP server is read-only by design and is the intended way in:
| question | tool |
|---|---|
| what does this bank hold? | list_facts(client) — the whole bank in one call, loads no models |
| one fact with its provenance | get_fact(qname); at= what the bank knew that day, world_at= what held in the world |
| you don't know the qname | search(query, client) — semantic, over facts and source documents |
| the source says something, but no fact exists yet | provgraf search-docs "…" over raw chunks |
| has a dilemma like this been resolved before? | precedents(query) — before resolving a new one |
| is the bank healthy? | check(client?) — before publishing anything built on these facts |
Chunks are not facts. doc_chunk holds raw document text for finding things; it carries no
approval and no staleness. Quote facts, not chunks. Promote a chunk to a fact only when a human
approves it.
Writing
Writes go through the CLI, where validation lives — never through raw SQL, which bypasses the invariants, and never through the MCP surface, which has no write tools.
provgraf add-doc <doc> --by <agent> --owner <slug> --file <path> # file-backed source
provgraf add-doc <doc> --by <agent> --owner <slug> --testimony --date … # someone vouched for it
provgraf add <qname> --value X --from <doc> --owner <slug> [--status to_confirm]
provgraf revise <qname> --value X --from <new-doc> --by <agent> --note "why"
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 · 69 lines · 887 tokens per session scan A 2193de6f5fa5
provgraf AGENTS.md is an instructions file published in the GitHub repository Inarus/provgraf (0 stars, last pushed 20d ago), licensed MIT. It adds 887 tokens to every session, about $0.0044 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 instructions, from other repositories
m_flow AGENTS.md
AGENTS.md instructions for FlowElement-xinliuyuansu/m_flow, covering m-flow — developer & agent reference, 1. repository map, extension points, 2. local development and python backend (requires python 3.10 – 3.13).
llm-wiki-agent GEMINI.md
Gemini CLI instructions for SamurAIGPT/llm-wiki-agent, covering llm wiki agent — schema & workflow instructions, how to use, directory layout, page format and ingest workflow.
obsidian-llm-wiki AGENTS.md
Instructions for green-dalii/obsidian-llm-wiki, covering llm wiki plugin project development standards, 🛡️ six-gate quality closure, gate 1: five-gate automated, gate 2: no side effects and gate 3: no breaking changes.
remnic AGENTS.md
Instructions for joshuaswarren/remnic, covering remnic - agent guide, architecture boundaries (non-negotiable), upstream references, adapter implementation rules and openclaw compatibility window.
LeanKG AGENTS.md
Instructions for FreePeak/LeanKG, covering leankg — agent context, build & test, cli quick reference, mandatory: docker mcp project paths and tool discovery prefer-order.
ontology-atlas AGENTS.md
Instructions for wlsdks/ontology-atlas, covering agents.md — ontology-atlas, product and non-negotiable architecture, start here, structure and routes and operating gates and skills.