@houserules/plugin-persona-auditor
v0.1.2
Published
houserules plugin: a blind-rank-then-reconcile persona-auditor agent template.
Maintainers
Readme
@houserules/plugin-persona-auditor
An agent asked to check whether a system's choice serves a given persona will read the selection code first if you let it, and then rationalize whatever that code already decided. That check confirms the engine's own logic instead of testing it against the persona.
This plugin ships a template for an agent that ranks blind before it looks: it forms its own ranking from a persona's stated priorities and the option data alone, only then reveals what the system actually chose, and buckets any divergence by a typed cause instead of guessing at it. It is a template you instantiate per component, not a module that installs standalone.
If your install carries a persona-auditor module id from before the CLI's built-in modules
moved into plugins, install this package to restore it under its plugin id.
Install
pnpm add -D @houserules/plugin-persona-auditor
pnpm exec houserules initRequires @houserules/cli.
init is what writes the module into .claude/. This module is off by default, so select it
when init asks.
Modules
persona-auditorinstalls.claude/templates/agents/persona-auditor.agent.md.template, a read-only, haiku-model agent template. Fill in<COMPONENT>,<PERSONA>, the priorities source, and the option data source, and instantiate one copy per component you want audited. The template forbids reading the selection or scoring code, ranks the options blind, then reveals the system's actual choice and reports one JSON verdict:AGREES,DEFENSIBLE_ALT,PRIORITY_UNDERWEIGHTED,CONSTRAINT_UNSTATED,OPTION_OVERLOOKED, orDATA_STALE.
Part of houserules
houserules is a portable set of Claude Code infrastructure that keeps the agent's context lean. This is one of twelve first-party plugins. The package list has the rest.
License
MIT. See LICENSE.
