Name: Evidence Integrity Gate
Price: 34 USD
Availability: InStock
Author: Agensi

See it in action

You say

Run the evidence integrity gate on this release note before I publish it. DRAFT: "Our new API eliminates latency and guarantees 100% uptime. It supports Node 16 and is 3x faster than v1." SOURCES: docs/api-reference.md, package.json, bench/results.json.

Your agent does

=== EVIDENCE INTEGRITY GATE === Draft: release-note.md | Sources provided: 3 | Files checked: 3

VERDICT: FAIL

BLOCKING ISSUES (must fix before ship):

UNSUPPORTED: "guarantees 100% uptime" — no source supports this -> remove or qualify
CRITICAL: "supports Node 16" — package.json engines requires >=18.0.0 -> fix version claim
Stage 1 SOURCE LEDGER: 3 sources (2 RELY, 1 CORROBORATE, 0 AVOID)
Stage 2 SUBSTANTIATION: 4 claims (1 GROUNDED, 1 INFERRED, 0 WEAK, 1 UNSUPPORTED)
Stage 3 VERIFICATION: 1 CRITICAL, 1 WARNING ("3x faster" cites no benchmark) -> FAIL
Stage 4 STRESS TEST: Weakest point: "eliminates latency" overclaim -> REVISE

TOP 3 FIXES, RANKED:

Correct the Node version to >=18 (CRITICAL mismatch).
Remove the 100% uptime guarantee or cite an SLA.
Soften "eliminates latency" to "reduces average latency" and cite bench/results.json for "3x faster."

What you get

Final gate before publishing a research summaryQuality gate in a multi-agent RAG pipelineAuditing AI-drafted docs and READMEs before merge

About this skill

Evidence Integrity Gate is the single control layer that runs your whole evidence suite as one ordered pre-ship gate. Drop in any AI-generated draft and it sequences four specialist checks, then returns one verdict.

What it does, in order: grades the sources (ranks every source by type, recency, authority, independence, and corroboration, and resolves conflicts by evidence weight); grounds every claim (labels each claim GROUNDED, INFERRED, WEAK, or UNSUPPORTED against a real source); verifies technical assertions against the actual files (versions, API references, performance figures, and code); and runs an adversarial stress test that hunts the weakest load-bearing claim, overclaims, and missing limitations. The output is one consolidated PASS / REVISE / FAIL report with ranked, blocking fixes.

Built for teams that cannot afford hallucinations: research summaries, docs and READMEs, market briefs, RAG answers, release notes, and multi-agent pipeline output. Install it as the last step before anything reaches a human, a customer, or a merge.

This gate orchestrates four standalone skills, also sold together as the Evidence Integrity Suite bundle. If you own them, each stage upgrades to its native, best-quality engine; if not, the gate runs proven inline fallback logic so you still get a real ship decision.