How to use Claude to build an RPG rules engine
Claude Code project agents built a standalone RPG evaluator with independent tests. Seeded scenarios exposed outdated rules; Foundry parity was not claimed.
tapestria.quest
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Use the supplied current SRD rules and reference data to build a standalone seeded Python evaluator. Separate test-writing and implementation permissions, prevent implementation agents from changing expectations, and require cited scenarios with expected event streams. Surface disagreements for review rather than using model memory as ground truth.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Tapestria
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- A standalone seeded rules engine shipped for Tapestria after schema changes and independent scenario tests. The author explicitly does not claim complete Foundry behavior parity.
- Human involvement
- The developer selected rules and data architecture, separated test and implementation permissions and required source citations and independent review.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- The developer separated reusable rules data from engines that depended on Discord or Foundry runtimes.
- Spec and implementation agents worked with separate edit permissions, followed by independent review.
- Seeded scenarios exposed missing and outdated rule behavior before the standalone engine shipped.
Tapestria's developer used coding agents to build Nat20, a standalone deterministic RPG rules engine. The project's origin account identifies Claude Code as its development harness; the later engine report describes separate agents and review by another model family.
The first rules implementation drifted because agents interpreted the source material independently. Reusing existing engines directly also failed: important logic depended on Discord or Foundry's runtime. The developer instead separated reusable data and design from the standalone Python evaluator.
For each effect type, one agent wrote tests from source rules and reference implementations, while another implemented behavior without permission to edit those tests. A separate reviewer checked the diff and blocked changes that made tests fit the implementation. Scenario fixtures supplied reproducible actions and expected event streams.
A later move to tactical map geometry required changing the schema and rewriting the evaluator again. Scenario work caught a placeholder spell-save calculation and repeated use of an older Counterspell rule from model memory. The shipped engine takes encounters and seeds and returns events without knowing the narrator. The author claims reproducible behavior for the game's needs, not complete parity with every rule in Foundry.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Developer blog · Tapestria
Frequently asked questions
Why not import an existing engine?
Candidate logic depended on surrounding Discord or Foundry runtime systems.
Could implementation agents change tests?
No. Test writing and implementation permissions were deliberately separated.
What level of parity is claimed?
Reproducible behavior needed by Tapestria, not complete Foundry parity.
More details
About this account
The developer separated reusable rules data from engines that depended on Discord or Foundry runtimes. Spec and implementation agents worked with separate edit permissions, followed by independent review.
Primary devlog, origin account and Nat20 repository opened. Claude Code attribution comes from the same project's origin account; later post describes multiple model families. No full Foundry parity is claimed. Suggested prompt written by Agents In Use; no original task prompt was published in the verified account.
Credit: Tapestria — Developer blog
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Python · Foundry SRD data · Scenario fixtures
Deterministic rules · Independent tests
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.