EricD1012

AI Research Datapack Protocol — Data skill for Claude Code

Data community

Claude skill for auditing, merging, and commissioning AI-generated research data packs.

How to install AI Research Datapack Protocol

This entry records only its repository, not the path inside it, so there is no exact command to give. Open EricD1012/ai-research-datapack-protocol and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Research Datapack Protocol does

Claude skill for auditing, merging, and commissioning AI-generated research data packs. Seven structural failure modes; negative judgments must be traced to source first.

Alternatives in Data

Financial Services — Reference agents, skills, and data connectors for the financial-services workflows we see most — investment ba 34.5k ★
Build Failure — Diagnose elasticsearch-js CI and local build failures 5.3k ★
Portaljs Add Dataset — Add a dataset (CSV, TSV, JSON, or GeoJSON) to an existing PortalJS portal 2.3k ★

README

AI Research Data-Pack Protocol

A Claude skill for **auditing, merging, and commissioning AI-generated research**.

When you delegate a market scan, a competitor teardown, or an event report to an AI agent, the output usually looks credible. That is exactly the problem. This skill treats every AI-produced data pack as *presumed broken until it passes seven structural gates*, and forbids the reviewer from rejecting any claim without first tracing it back to its original source.

Distilled from a four-agent bake-off on China's 618 shopping festival (beauty and personal care, June 2026), where four different agents produced the same research brief and were cross-evaluated.

What it does

Three scenarios, chosen at Step 0:

Scenario	Input	Output
Commission	"I want an agent to research X. How do I write the brief?"	An eight-section brief template with hard epistemic rules baked in
Evaluate	One or more AI-generated research packs	Each pack scored on two independent axes (credibility, usefulness), plus a list of which of the seven failure modes it hit, with sourced evidence for every rejection
Merge	Several packs on the same topic	One consolidated version, assembled layer by layer (narrative skeleton, numbers, methodology) and swept through all seven gates last

The seven structural failure modes

These are category-agnostic. They show up in AI research about any market, platform, or event.

Year contamination — a headline number from a previous cycle is presented as current.
Metric conflation — counts and revenue, or platform-wide and category-level figures, are placed side by side or added together.
Period mismatch — year-over-year comparisons across windows of different lengths.
Interim value as final — a mid-event snapshot is reported as the closing result.
Cross-wave leaderboard merging — rankings from different phases are stitched into one "overall" list.
**Precisi