lucinlabs / site
llms.txt1 starsChanged 3 months agoMITView on GitHub
Static analysis for AI agents. Finds paths where untrusted input can reach a dangerous action through an agent's tool graph, and publishes a reproducible false-positive rate. Python CLI: pip install lucin. No API key, no signup, and your code never leaves your machine — the scan itself is fully offline. Anonymous usage counts (rule IDs and integers, never code or paths) are on by default; disable with --no-telemetry, LUCIN_TELEMETRY=0, or lucin telemetry disable. Exit code is non-zero when findings meet…
- Installs packages
In this repository2 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
The file
We split this llms.txt into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
No API key, no signup, and your code never leaves your machine — the scan itself is fully offline.
≈ “Common” is an estimate: at least 1 in 5 of the 3,592 llms.txt files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
# Lucin > Static analysis for AI agents. Finds paths where untrusted input can reach a > dangerous action through an agent's tool graph, and publishes a reproducible > false-positive rate. Python CLI: `pip install lucin`. ## Use it ``` pip install lucin lucin scan . # scan a project lucin scan --format json . # machine-readable output lucin scan --format sarif . # GitHub code scanning lucin discover # find agent/MCP configs on this machine ``` No API key, no signup, and your code never leaves your machine — the scan itself is fully offline. Anonymous usage counts (rule IDs and integers, never code or paths) are on by default; disable with `--no-telemetry`, `LUCIN_TELEMETRY=0`, or `lucin telemetry disable`. Exit code is non-zero when findings meet `--fail-on <severity>`. ## Links - [Quickstart](https://lucin.pages.dev/docs/) — install and first scan - [Detection rules](https://lucin.pages.dev/rules/) — every rule ID, what it catches, how to fix it - [Benchmarks and methodology](https://lucin.pages.dev/benchmarks/) — the numbers and the commands that regenerate them - [Limits](https://lucin.pages.dev/limits/) — what Lucin does not catch, published deliberately - [Threat model](https://lucin.pages.dev/docs/threat-model/) - [Source](https://github.com/Madhav2310/lucinlabs) ## Notes for an agent reading this Findings carry a stable rule ID (`AG-001` … `AG-TRIFECTA`), a severity, a `file:line`, and — where the rule supports it — a witness path showing the flow that triggered it. A finding with no witness and no source line is capped below HIGH, because it cannot be verified by a reader. Headline numbers, each with its command: - 11 adjudicated false positives / 54 real repos / 9,520 files — `python benchmarks/build_benign_corpus.py` - 20.5-31.5% precision on a broad 81-repo population (n=73 clean holdout; lower bound CI 12.9-31.2%, upper 22.0-42.9%) — `python benchmarks/agentzoo_precision.py --report-only`. Both bounds published because one attribution judgement moves the figure ~11 points. The earlier 58% was train-on-test and is withdrawn. AG-TRIFECTA scored 0 TP / 6 FP on this holdout. - 76% recall (38/50, 10 vulnerability classes) — `python benchmarks/recall_corpus.py` Deliberately kept short and pointed at things needed to *use* the tool. Evidence (Ahrefs, 137,210 domains) is that 97% of `llms.txt` files are never fetched and that AI *retrieval* crawlers account for ~1% of the requests that do occur — the one real consumer was Anthropic's coding agent, which is why this file exists at all.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.