dual-source-honesty

Keeps fetch-time catalogs distinct from corrected bibliographies and labels numeric values by provenance status.

Updated Sep 2, 2026
One-click install
npx skills add https://github.com/ricfulop/cba-agentic-engineering-bootstrap --skill dual-source-honesty-ricfulop
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: dual-source-honesty
Source: https://github.com/ricfulop/cba-agentic-engineering-bootstrap/tree/main/skills/dual-source-honesty
Command: npx skills add https://github.com/ricfulop/cba-agentic-engineering-bootstrap --skill dual-source-honesty-ricfulop

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When cataloging papers or recording numbers, two sources often disagree, and silently copying one over the other corrupts the record. This Skill enforces rules that keep raw fetch results, corrected bibliographies, and value provenance statuses separate so disagreements become explicit decisions instead of silent merges. ## Core Features & Use Cases - Literature separation: Preserves the fetch-time catalog (catalog.json) as-is while routing bibliography corrections into refs.bib plus a register, keeping empty author lists empty and unconfirmed truncations as "and others". - Number provenance statuses: Enforces a status vocabulary (MEASURED, DERIVED, CATALOG, ESTIMATE, blank/NR/TBD) so workbook values are never inferred from adjacent cells or unrelated columns. - Use Case: While building a literature corpus, a search returns a paper with a truncated three-author record; instead of inventing authors, the record stays "and others" and the correction is logged in refs.bib with a register entry. ## Quick Start Apply the dual-source-honesty rules when cataloging these search results and recording the measured values in my workbook.

Frequently Asked Questions about dual-source-honesty

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I handle disagreements between a search catalog and my bibliography?

Leave the fetch-time catalog (catalog.json) unchanged and record corrections in refs.bib plus a register. A disagreement is treated as a decision request, never as permission to copy one source over the other.

What should I do with empty or truncated author lists in paper records?

Empty author lists stay empty rather than being invented. Truncated three-author records that were not confirmed remain marked as "and others", and a catalog count is never treated as a reading count.

What do MEASURED, DERIVED, CATALOG, and ESTIMATE mean in a workbook?

MEASURED means an instrument or run produced the value, DERIVED means it was computed from named existing inputs, CATALOG means it came from a vendor or index field, and ESTIMATE means judgment, ideally bounded. Blank, NR, TBD, or PLACEHOLDER mean unknown.

Can I fill unknown table cells from adjacent columns or manuscript prose?

No. Blank, NR, TBD, and PLACEHOLDER cells stay unknown and must not be filled from adjacent cells. Procurement metadata cannot be inferred from a model-spec column, nor a canon value from manuscript prose.

When is private correspondence quotable in research writing?

Private correspondence pasted into chat is background material, not quotable content. It only becomes quotable when the author explicitly clears the specific quote for use.