Works with the AI Agents you already use
Claim Receipt Gate
38% becomes 45%, 2.1x becomes 3x, and "tests pass" turns into "independently audited".
$12
Claim Receipt Gate
Example session with this skill installed
Here's the v2.4 launch post my agent wrote (draft.md) and our evidence.json from the benchmark and CI run. Can you check it before I post it?
- Read your context and instructions
- Compiled the claim receipt gate
I ran claim-receipt-gate on draft.md against evidence.json. Verdict: REFUSE (exit code 1): 7 of 9 checkable claims are not backed.
{
"kind": "claim-receipt-gate",
"verdict": "REFUSE",
"why": "7 unbacked claim(s): https://example.com/beta, 214/214, 45%, 8%, $50k, 3x, audited",
"checked": 9,
"unbacked": [
{ "type": "score", "words": "214/214", "line": 3, "why": "claims 214/214 but no evidence backs that exact score" },
{ "type": "verified", "words": "audited", "line": 3, "why": "says \"audited\" but the evidence does not mark it verified" },
{ "type": "pct", "words": "45%", "line": 5, "why": "claims 45% but no fact carries that percentage" },
{ "type": "multiplier", "words": "3x", "line": 6, "why": "claims 3x but no fact carries that multiplier" },
{ "type": "money", "words": "$50k", "line": 7, "why": "claims $50k but no fact carries that amount in that currency" },
{ "type": "pct", "words": "8%", "line": 8, "why": "claims 8% but no fact carries that percentage" },
{ "type": "url", "words": "https://example.com/beta", "line": 10, "why": "cites a url the evidence does not list (not the live url, not in links)" }
],
"backed (trimmed)": [
{ "type": "live", "words": "live", "line": 3 },
{ "type": "ci", "words": "tests passing", "line": 3 }
],
"draftSha256": "7ab7ada5e10385a4fe7baa7e50e6a8c501dca982223dc3ab259674bbf19ce50f",
"evidenceSha256": "9f39b029b3065b3c7fac80ee8cb6a9e19d708f71a3ad8aaa6d7c9b7faf320390",
"reportSha256": "bcbde856dc4292acff3f302cbbc9e877fc0527fa32faecdb37eecfc95054d27c"
}
What happened
- The tests were 212/214, not 214/214, and nothing says the release was independently audited.
- The benchmark shows p95 down 38% and cold starts 2.1x faster; the draft says 45% and 3x.
- "$50k saved" and "8% on memory" have no measurement behind them at all.
- https://example.com/beta isn't the live URL or a listed link.
I'd change the figures to 212/214, 38% and 2.1x, drop "audited", "$50k" and "8%" unless you have numbers for them, and use the changelog URL. Then run the gate again for a SHIP seal.
Connects securely to your tools. The creator never sees your data.
About this skill
Agent Trust: every number in the post needs a receipt
AI-written release notes and launch posts round up. 38% becomes 45%, 2.1x becomes 3x, and "tests pass" turns into "independently audited". This skill reads the draft, pulls out every claim it can check, and holds each one against the evidence you actually measured. If any claim has no backing, the draft is refused, with the words and the line.
What it does
From any text or Markdown draft it extracts:
- Scores such as 212/214 (dates like 31/10/2026 are ignored)
- Percentages, money amounts in $, £ or € (including k, m, bn) and multipliers such as 2.1x
- URLs, which must be the recorded live URL or a listed link
- Live / shipped / deployed, CI green / tests passing, and verified / audited wording
Each is checked against a JSON evidence file: url, live, ci, tests {passed, total}, verified, links and free-form facts. A number is only backed when a fact carries that exact number on its own, so "5%" is not backed by "15%", and £860 is not backed by $860.
When an agent should use it
- Before you publish anything an agent wrote that cites results
- Release notes, changelogs, launch posts, investor or client updates, READMEs
- When you want a record of exactly what was claimed and what backed it
What it outputs
JSON: SHIP or REFUSE, every claim with type, words, line and why it is or isn't backed, and SHA-256 hashes of the draft, evidence and report. A SHIP comes with a seal binding the draft to its evidence; verify-seal shows if either changed. Optional hash-chained ledger of every decision. Exit 0 = SHIP, 1 = REFUSE, 2 = bad evidence.
Limitations
- It checks that the draft doesn't claim more than the evidence. It can't tell if the evidence itself is true, so fill it from real measurements.
- Claims written as plain words ("doubled", "most users") are not detected. Bare numbers that are not scores, percentages, amounts or multipliers are left alone on purpose.
- English wording only for the live/CI/verified checks.
Open source, packaged skill
Original work by Simon Gant. The core logic is public under the MIT licence (post-proof). This listing is the packaged, tested agent skill: SKILL.md workflow, CLI, general evidence format, money and multiplier claims, stricter number matching, seal and ledger, and docs. Requires Node.js 18+, zero dependencies, no network, no API keys.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed today
- Software
Node.js 18+
Permissions
Runs one local Node.js script (`node scripts/claim-receipt-gate.js`) through the agent's shell tool. Reads only the draft and evidence files you pass it. Writes a file only when you pass `--ledger` (the ledger you name). No network access, no environment variables read. Pure compute.