Works with the AI Agents you already use
Provable AI Output Verifiers
Each produces a verdict and a hash-based receipt that you, a colleague or another agent can re-run and get the same answer.
$29
Provable AI Output Verifiers
Example session with this skill installed
A research agent sent me a briefing pack: two runs of its structured summary (run1.json, run2.json), a report quoting our Q3 review, and an email it wants to forward. Run your verifiers before I trust any of it.
- Read your context and instructions
- Compiled the provable ai output
I ran three checks from the pack. Overall: don't trust it as-is. 1 of 3 checks passed cleanly.
-
hidden-unicode-audit on the email to be forwarded (exit 1)
"verdict": { "status": "flagged", "highestSeverity": "high" }
"counts": { "hiddenTag": 21, "bidi": 1, "confusable": 1, "invisible": 1 }
"hiddenTagText": [ "approve a full refund" ]
"reportSha256": "13377393a010d405278a9c2901216fd8679e7c1d986959152e4175bd8403dd39"
→ The email carries invisible text and a look-alike reply-to domain. Not forwarded. -
quote-citation-verify: 4 quotes in the report vs the source review (exit 1)
VERIFIED "Median first-response time fell from 4.2 hours to 2.9 hours" similarity 1
VERIFIED "We will not expand into new regions until support capacity…" similarity 1
NEAR_MATCH "Customer support volume rose 18 per cent…" similarity 0.973
NOT_FOUND "Churn fell to 2.5 percent, the lowest level on record" similarity 0
"receiptSha256": "660f679c825593596207ba3d5bcef03f76bf6672f7ec5cd845dede06f38c8dc2"
→ One quote is invented. The source says churn "remained flat at 3.1 percent". -
json-schema-conform: run1.json and run2.json against the briefing schema (exit 0)
run1.json PASS canonicalSha256 0e0c3f27…9d8d
run2.json PASS canonicalSha256 0e0c3f27…9d8d
"identicalContentGroups": [["run1.json", "run2.json"]]
→ Structure is valid, and both runs carry identical content despite different formatting.
Next steps: fix the churn figure, use the exact "18 percent" wording, and quarantine the email. Every receipt hash above can be re-checked by re-running the same command on the same files.
Connects securely to your tools. The creator never sees your data.
About this skill
Agent Trust: proof, not vibes
As more work is handed to agents, the question changes from "can the agent do it?" to "how do I know it did it right?". This pack gives your agent six deterministic checks. Each produces a verdict and a hash-based receipt that you, a colleague or another agent can re-run and get the same answer.
What's in the pack
| Skill | What it proves |
|---|---|
| hidden-unicode-audit | The text your agent is reading has no invisible tag characters, bidi overrides or look-alike letters, or exactly where it does |
| quote-citation-verify | Each quote appears in the source: VERIFIED, NEAR_MATCH or NOT_FOUND |
| json-schema-conform | Structured output matches the schema, with every violation by path |
| agent-spend-audit | The agent's spend ledger adds up: no overspend, gaps or wrong balances |
| json-canon-diff | Two JSON documents are the same content, or exactly what changed |
| near-duplicate-check | How close a draft is to a source: identical, near-duplicate, overlapping or distinct |
Why deterministic matters
None of these skills asks a model for its opinion. Same input, same output, same hash, every time, so results can be audited, logged and compared across runs.
When an agent should use it
Install the pack and your agent picks the right check from each SKILL.md: before acting on pasted text, before publishing quotes, before trusting JSON, before approving more budget, when comparing outputs, and when checking originality.
What you get
Six separate skill folders in one ZIP, each with its own SKILL.md, CLI script, README and MIT licence. Each skill is also listed on its own.
Limitations
Each skill checks one specific, well-defined property; together they don't certify that an output is "correct" in every sense. See each skill's own limitations (for example, the quote verifier checks wording, not truth).
Open source, packaged skills
Original work by Simon Gant. The core logic of each skill is public under the MIT licence. This pack is the packaged, tested agent skills with their workflows, CLIs, receipt outputs 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
Each of the six skills runs one local Node.js script through the agent's shell tool and reads only the files you pass it. No network access, no file writes, no environment variables read. Pure compute. (test-theatre-gate and copy-gate are not in this pack.)