Trade crypto, manage a multi-chain wallet, and query an AI analyst from one CLI.
Security
Ed25519 Signature Verifier — CCS Audit Receipts
Try itOffline Ed25519 signature verification for AI-agent audit receipts. Verify that a CCS receipt was signed by a known signer and never tampered with: Ed25519 verification, RFC 8785 JCS canonicalization, SHA-256 content-hash recomputation and 22-field schema/tamper checks, using the vendored open-source CCS verification core. Zero network calls; only the receipt and the issuer public key are needed, private keys are never involved. Use when verifying a CCS receipt, verifying an Ed25519-signed agent decision receipt, or proving whether a signed agent-tool-call receipt was tampered after issuance.
What it does
CCS Receipt Verify — Ed25519 local receipt verification
The skill document
CCS Receipt Verify — Ed25519 local receipt verification
Don't trust the receipt. Verify it.
When you hold a CCS (Correctover Conformance Shape) AI-agent audit receipt and need to independently confirm "it was really signed by the issuer, and no field changed after signing", use this skill. Verification runs locally by default (vendored open-source CCS verification core, no network, receipt data stays on the machine). You only supply the receipt itself and the issuer's public key; the private key is never needed and never sent.
中文:本地验证单张 CCS AI Agent 审计收据——Ed25519 验签、RFC 8785 JCS 规范化、SHA-256 摘要重算、22 字段 schema 校验与篡改证据,默认零网络、 数据不出本机,只需收据和签发方公钥,私钥永不参与。
What it does
A full verification pipeline on a single CCS receipt (local by default):
- RFC 8785 JCS canonicalization — deterministic JSON serialization of
the 20 fields excluding
signature/content_hash(keys sorted by UTF-16, non-ASCII lowercased\uXXXX, ECMAScript shortest-number representation). - SHA-256 content-hash recomputation — recompute the digest and compare
against the receipt's embedded
content_hash. - Ed25519 signature verification — verify the canonical payload against the issuer public key you provide (PEM/DER/raw 32 bytes all accepted).
- 22-field schema check — field completeness, types, enums
(
allow/block), the 7checksdimension statuses, base64 signature length. - Tamper evidence — on signature or hash failure, emit a cryptographic proof list, the signed field scope (20 field names), and internal consistency hints.
Stateless and idempotent: verifying the same receipt twice yields byte-identical results.
What it does NOT do
- It does not issue receipts and does not run the agent runtime 7-dimension checks (that is the CCS runtime's job).
- It makes no policy judgment: whether a receipt "should have allowed or blocked" is a policy question; this skill only answers "was it altered after signing / signed by the holder of the matching private key".
- It never needs or receives any private key.
- Local mode logs nothing remotely.
Usage
# Verify a receipt (human-readable report)
python3 scripts/verify_receipt_online.py \
-r examples/sample_receipt.json \
-k examples/sample_pub.pem
# Machine-readable JSON
python3 scripts/verify_receipt_online.py \
-r receipt.json -k pub.pem --json
# Read receipt from stdin
cat receipt.json | python3 scripts/verify_receipt_online.py -k pub.pem
Exit codes: 0 = receipt valid; 1 = invalid/tampered (report still
printed — this is a business result, not a crash); 2 = input error.
Arguments:
| Arg | Description |
|---|---|
-r/--receipt | Receipt JSON file; omit or - for stdin |
-k/--public-key | Issuer Ed25519 public key PEM file (required) |
--json | Emit the full JSON report |
--online | Explicitly use Correctover's hosted verification (default is local, offline) |
--endpoint | Online mode: override the service endpoint (for self-hosted deployments) |
Privacy boundary: default verification is entirely local — the script
makes no network request, and neither the receipt nor the public key leaves
the machine. The verification core is the open-source CCS implementation
(same code as the official hosted service). Only with an explicit
--online flag are the receipt and the issuer public key sent over
HTTPS to the verification service; the private key is never needed, read,
or transmitted in any mode.
Report fields
{
"valid": true,
"errors": [],
"evidence": {
"schema": {"ok": true, "issues": []},
"jcs": {"canonical_bytes": 1247, "digest_sha256": "9dff2be4…"},
"content_hash": {"match": true, "claimed": "…", "recomputed": "…"},
"ed25519": {"status": "valid", "verified": true}
},
"tamper": null,
"receipt_ref": {"decision_id": "…", "agent_id": "…", "verdict": "allow"}
}
On tampering, valid=false, tamper.detected=true, tamper.proof lists
the cryptographic proofs (signature failure / digest mismatch), and
tamper.signed_scope lists the 20 signed fields for field-by-field
comparison.
Example
[PASS] JCS canonicalization 1247 bytes, sha256=9dff2be41aa2d14c…
[PASS] content_hash match
[PASS] Ed25519 verification status=valid
[PASS] 22-field schema check
RESULT: PASS: receipt authentic and valid (4.1 ms)
Tampered sample (examples/tampered_receipt.json — verdict and
checks.security.status altered, signature unchanged):
[FAIL] content_hash match
[FAIL] Ed25519 verification status=invalid
⚠ Tamper evidence:
- Ed25519 signature verification failed over the JCS canonical payload
- content_hash does not match recomputed digest: signed fields changed after issuance
RESULT: FAIL: receipt invalid / tampered
Requirements
Python 3.7+, standard library only (urllib/json/hashlib). No
third-party packages required for local mode.
About
By Correctover (Guigui Wang) — Correctover = AI Reliability. CCS (Correctover Conformance Shape) is an open, vendor-neutral runtime verification standard for AI agents: 7 dimensions (Structure/Schema/ Latency/Cost/Identity/Integrity/Security) with Ed25519-signed receipts that turn "the agent did it" into auditable, tamper-evident evidence. Homepage:
Disclaimer
Results are auto-generated and apply only to the submitted input. Automated verification is an aid only and constitutes no security guarantee, compliance conclusion, audit opinion, or legal opinion. No endorsement by OWASP, IETF, or any standards body is claimed. The business impact of a tampered receipt is outside the scope of automated checks; security decisions must ultimately rely on qualified human review.
License
MIT-0 for this ClawHub distribution. Anyone may use, modify, and redistribute it, including commercially, with no attribution required.
Related skills
Join a video meeting as an AI bot with voice, avatar, and screenshare across four operating modes.
Generate and edit Draw.io, Mermaid, and Excalidraw diagrams from natural language using a structured JSON spec.
Stores durable facts in a categorized, plain-markdown vault on disk, alongside your agent's built-in memory.
Query and manage Linear issues, projects, teams, cycles, labels, and comments through a managed OAuth GraphQL endpoint.
Manage Stripe customers, subscriptions, invoices, products, prices, and payments through OAuth-authenticated API calls.
More from dshcorrectover
Browse all skillsMCP security audit for AI-agent configs — run it before connecting any MCP server. 14 automated static rules catch hardcoded credentials, arbitrary command execution, writable mounts, SSRF attack surface, prompt-injection risks, plaintext HTTP endpoints and unpinned remote launchers, so unsafe agent tool calls get stopped before they run. Outputs CRIT/HIGH/MED/LOW findings with concrete fix advice. Runs 100% locally with zero network calls. Use when reviewing, auditing or scanning an MCP/agent server config (mcpServers JSON) for security issues, before wiring it into an agent.
Batch Ed25519 signature verification for up to 200 AI-agent audit receipts, fully offline. Every receipt is checked against one issuer public key with Ed25519 verification plus RFC 8785 JCS/SHA-256 integrity checks; outputs aggregate valid/invalid statistics, pass rate, tampered-index list with per-receipt failure reasons, and optional hash-chain linkage, using the vendored open-source CCS verification core. Zero network calls. Use when batch-verifying CCS receipts, auditing a batch of signed agent decision receipts, or verifying a receipt chain.