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/automators-com/flowproof/claude-mdgit clone --depth 1 https://github.com/automators-com/flowproofWhat 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 | $0.01368 | $0.01368 |
| Opus 5 | $0.00684 | $0.00684 |
| Sonnet 5 | $0.00274 | $0.00274 |
| Haiku 4.5 | $0.00137 | $0.00137 |
Grade A, and why
flowproof 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 2d 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 — 124 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
Guidance for Claude Code (claude.ai/code) working in this repository.
If you are an autonomous loop, read CHARTER.md first. It is
the direction, the invariants, and the out-of-scope list, and it outranks
anything here. Your role prompt is in scripts/loop/roles/. This file is
context; the charter is authority.
What flowproof is
Test an AI agent the way you test everything else: run it once, keep the recording, assert against it from then on.
flowproof sits at the model boundary. It captures a real run — every model request and every tool-call decision — into a trace, then serves that recording back on later runs. Replay makes zero LLM calls, so a suite that cost money per commit and flaked on sampling becomes free and repeatable.
What is asserted is behaviour, not prose: which tools were called, with which arguments, in which order, and which were not.
Repository layout
Rust workspace. All crates move together on one version.
| Crate | What it is |
|---|---|
flowproof-trace |
trace format, selector ladder, trace-to-script compiler |
flowproof-replay |
deterministic executor — replays traces with zero LLM calls |
flowproof-driver |
native driver: screen capture, input injection, UI Automation |
flowproof-agent |
recording agent: planner loop, pluggable model backends |
flowproof-adapters |
SAP GUI Scripting COM, browser via CDP (behind feature flags) |
flowproof-cli |
the flowproof binary: record, run, heal |
flowproof-python |
PyO3 bindings — the flowproof._native extension module |
Also: sdk/python (Python SDK, uv/ruff/pytest), sdk/js (placeholder),
docs/ (design + format), examples/, tests/flowproof/ (flows exercising
flowproof itself), scripts/gate/ and scripts/loop/ (the autonomous-loop
constitution — see below).
The driver is Windows-native, but the workspace must always build on Linux and macOS via the stub backend. Assume your change will be compiled on Linux.
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.
- 2d ago First seen · 124 lines · 1,368 tokens per session scan A cf2c64b97034
flowproof CLAUDE.md is an instructions file published in the GitHub repository automators-com/flowproof (5 stars, last pushed 2d ago), licensed Apache-2.0. It adds 1,368 tokens to every session, about $0.0068 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
keploy AGENTS.md
Instructions for keploy/keploy, covering agents.md, what this project is, build, run, lint, test, build the binary and standard build (what ci calls "build-no-race").
playwright-java-test-automation-architecture GEMINI.md
Instructions for Tahanima/playwright-java-test-automation-architecture, covering project: playwright java test automation architecture, context & tech stack, project mapping, development rules and locator strategy (priority order).
selenium-boot CLAUDE.md
Claude Code instructions for seleniumboot/selenium-boot, covering claude.md, project overview, common commands, run all unit tests and full build (compile + test + package).
claude-code-discord AGENTS.md
Instructions for zebbern/claude-code-discord, covering agents guidelines, presentation & ux, knowledge freshness & research preferences, testing & verification preferences and real runs over green mocks.
jumpstarter CLAUDE.md
Instructions for jumpstarter-dev/jumpstarter, covering claude ai instructions, project rules and guidelines, when to read these rules and key information.
datamimic AGENTS.md
Instructions for rapiddweller/datamimic, covering agents.md, when should an agent reach for datamimic?, tool selection, benchmarking local ollama models and authoring a new model.