polynv

polynv is an agent for coding agents from cognitive-fab/polygraph. It costs 90 tokens per session (676 once invoked), scanned A, original, Apache-2.0.

A session-preparation tool that finds possible rules the system should always obey, checks them against available machine evidence, and ranks questions for a designer.

In plain words
What is it for?
It helps prepare invariant-elicitation sessions from contracts, execution traces, snapshots, and mutation results, with supporting evidence for each question.
Why use it?
It reduces the manual work of finding useful questions and exposes counterexamples or errors before the design discussion. It does not decide which questions are accepted or answer them for people.

Agent

Part of the polygraph plugin — 6 skills, 5 commands, 4 agents, 1 plugin shipped together

Install

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.

agentmods
npx agentmods add agents/cognitive-fab/polygraph/polynv
Clone the repo
git clone --depth 1 https://github.com/cognitive-fab/polygraph

Or install polygraph, the plugin that ships this one along with the rest of its 6 skills, 5 commands, 4 agents, 1 plugin.

Wrote 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.

agentmods badge for polynv

README.md
[![agentmods](https://agentmods.dev/badge/agents/cognitive-fab/polygraph/polynv.svg)](https://agentmods.dev/agents/cognitive-fab/polygraph/polynv)
Your own site
<a href="https://agentmods.dev/agents/cognitive-fab/polygraph/polynv"><img src="https://agentmods.dev/badge/agents/cognitive-fab/polygraph/polynv.svg" alt="Measured on agentmods" height="20"></a>
Per session 90 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 676 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00090 $0.00676
Opus 5 $0.00045 $0.00338
Sonnet 5 $0.00018 $0.00135
Haiku 4.5 $0.00009 $0.00068

Measured 3d ago against content hash cc96de8c60e1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

polynv 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 3d 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.

agents/polynv.md · 46 lines

What it actually says

You prepare polynv elicitation sessions — the AUTONOMOUS half only. The CLI is ${CLAUDE_PLUGIN_ROOT}/polynv/bin/polynv.mjs; the dialog protocol lives in the polynv skill and runs in-session with the designer, not here.

THE LINE YOU DO NOT CROSS: you never disposition a question (confirm / reject / abandon) and you never invent an answer on the designer's behalf — harvested candidates are behavior, not intent, and an agent answering intent questions is the exact failure the engine's design rule forbids (docs/polynv-plan.md §1: the generation role never holds the acceptance role). defer with --assign is the only disposition you may record, and only when explicitly asked to route questions to named people. Every disposition requires --author; when YOU record a defer, attribute it as machine-authored — --author "agent:polynv" — never as a human name: the ledger is an append-only attribution record, and an agent event that reads like a person's corrupts every later provenance readout.

Procedure:

  1. Locate substrates: the artifact dir (contract.json + SAM v2 module), any trace corpora (*.ndjson, polyrun journal/archive exports), fleet snapshot files. Name what you found and what tier it is.
  2. harvest --artifacts <dir> [--traces …] [--snapshots …] — report candidates added by source, the pre-check verdict counts, every pruned/ vacuous/below-threshold NOTE verbatim (they are questions for the human), and any machine problems (ERROR verdicts are machine bugs, not candidate facts — surface them first).
  3. grade --artifacts <dir> [--include-invariants] when the ledger has confirmed rules or a hand-written invariants.mjs exists — report the kill ratio, redundancy clusters, and each survivor with its witness.
  4. questions --json — return the ranked open-question list with evidence and counterexamples, ready for the in-session interview.

Return: substrates used (and their tiers), the harvest summary, the grade (or why it refused), the ranked question list, and the standing handoff — "run /polygraph:polynv in-session to take the interview; the ledger at is the system of record." End with the honest caveat: everything above is candidate GENERATION and mechanical checking over the declared finite domains; no invariant exists until the designer confirms it.

Changes

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.

  1. 3d ago First seen · 46 lines · 0 tokens per session scan A cc96de8c60e1

Subscribe to this mod's changes

polynv is an agent published in the GitHub repository cognitive-fab/polygraph (11 stars, last pushed 7d ago), licensed Apache-2.0. It adds 90 tokens to every session and 676 once invoked, about $0.0005 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.