@piposlabs/praxa-verify

Offline verifier for Praxa audit-trail hash chains. Zero runtime dependencies. Single file. ~150 lines of audited code.

If you use Praxa to log your AI agent compliance evidence, this is the tool that lets you prove Praxa never tampered with your audit trail — without trusting Praxa's servers. Export your chain, run praxa-verify, get an independent cryptographic confirmation.

Why this exists

Compliance auditors (SOC 2, EU AI Act, NIST AI RMF) increasingly want evidence that AI agent decisions are immutable. Praxa achieves this via a SHA-256 hash chain: every event references the hash of the prior event, so any tampering anywhere in the chain breaks all hashes from that point forward.

The chain is only as trustworthy as the code that verifies it. Rather than ask auditors to take our word for it, we publish the verifier as MIT-licensed zero-dependency code. Read it. Audit it. Run it on your own exported events.

Install

npm install -g @piposlabs/praxa-verify
# or invoke without install
npx @piposlabs/praxa-verify chain-export.json

Usage

CLI

# Export your chain from Praxa (https://praxa.piposlab.com/dashboard/agents/<id>/export)
# then verify locally:
praxa-verify chain-export.json

# Pipe from stdin
curl -H "Authorization: Bearer $PRAXA_KEY" \
  https://praxa.piposlab.com/api/v1/agents/<id>/export \
  | praxa-verify --stdin

Exit codes:

| Code | Meaning | | ---- | -------------------------------- | | 0 | Chain verified | | 1 | Tampering detected | | 2 | Usage / file error | | 3 | JSON parse error or wrong shape |

Library

import { verifyChain } from "@piposlabs/praxa-verify";

const result = verifyChain(events);
if (result.ok) {
  console.log(`Chain OK — ${result.chain_length} events, head=${result.head_hash}`);
} else {
  console.error(`Tampered at position ${result.tampered_position}: ${result.reason}`);
  process.exit(1);
}

Algorithm (v1, locked 2026-05-19)

For each event in chronological order:

entry_hash = SHA-256( canonicalize({
  payload: { ...event_without_previous_hash_and_entry_hash },
  previous_hash: <parent_entry_hash>  or  ""  at chain root
}) )

Where canonicalize is deterministic JSON serialization with:

• Object keys sorted alphabetically at every nesting level
• Arrays preserved in insertion order
• undefined values dropped, null values kept literal

A chain is valid iff:

1. Position 0 has chain_position=0 and previous_hash=null|""
2. Every position N's previous_hash equals position N-1's entry_hash
3. Every position N's entry_hash matches the recomputation from its payload

Known-answer test vector

If your build of praxa-verify doesn't reproduce this exact hash, something is wrong — either with your install or with this code. Don't trust it.

Input payload:
{
  "occurred_at": "2026-01-01T00:00:00.000Z",
  "model": "claude-test",
  "input_payload": { "a": 1, "b": 2 },
  "output_payload": { "ok": true },
  "triggered_by": "test",
  "downstream_effects": [],
  "chain_position": 0
}
previous_hash: null

Output entry_hash:
376f167cc23d397e93d7598cd259d4fc831ef31013eba59ee4cfd63417cb1b1b

Run praxa-verify fixtures/known-answer.json after install to confirm this verifier reproduces the exact same hash on your machine.

What about Praxa itself?

praxa-verify is the verifier. The full Praxa platform (auth, billing, dashboards, retention, webhooks) lives at praxa.piposlab.com — that part is closed source. The hash chain itself is described in our public algorithm spec and verified by the code in this package.

License

MIT — see LICENSE.

Pipo Lab LLC · [email protected]