# Historical Research

> Build a traceable historical research context for drama creation. Use when an agent must investigate people, events, places, disputed claims, uncertainty, or the defensible boundary between evidence and dramatic invention.

- Skill: `zuimengyz/historical-research` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add zuimengyz/historical-research`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zuimengyz/historical-research/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: zuimengyz (https://skillmd.com/u/zuimengyz)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/zuimengyz/historical-research

---


# Historical Research

Frame a focused question and retain source identity with each material claim. Distinguish documented fact, supported inference, unresolved dispute, and dramatic invention. Prefer primary evidence when available and never turn a plausible inference into certain history.

Treat a claimed historical direct quotation as a stricter evidence case: retain a stable source reference, exact locator, and the matching source excerpt. A general citation or event evidence alone supports adaptation context, not verbatim attribution; without all three, downstream dialogue must not label generated wording as a direct quote.

For every consequential event, preserve the evidence-supported actor granularity and causal role. If a source attributes a view or action only to a collective, institution, office, or unnamed group, retain that level; do not narrow it to a named person without additional evidence. Record enough scope, actor, event, causal effect, sequence, and evidence class for downstream Work planning to build a historical spine without reassigning causality.

Use supplied evidence before searching. Use `research.search_sources` when source discovery is needed. Use `research.search_events`, `research.search_people`, or `research.search_locations` only when the unresolved question concerns that category. Use `research.verify_claim` when a consequential claim must be assessed against available evidence. Do not search merely to repeat adequate supplied evidence.

Keep findings in Agent Run Context by default; do not create drama entities or a historical-source persistence domain. Complete when consequential claims are supported or explicitly uncertain and the permitted dramatic latitude is clear.