---
title: "rolecasting"
description: "Rolecasting: portable model selection and delegation topology for agent work."
canonical: https://agentpluginsdirectory.com/plugins/rolecasting--nisavid
last-updated: 2026-10-21
---

# rolecasting
Rolecasting: portable model selection and delegation topology for agent work.
- Slug: rolecasting--nisavid
- Publisher: Ivan D Vasin
- Repository: https://github.com/nisavid/provingkit
- Manifest: plugins/rolecasting/plugin.json
- Version: 1.0.0
- License: MIT
- Category (editorial): agent-tooling
- Skills: 2 (choosing-agent-models, delegating-cross-agent-work)
- MCP servers: 0
- Stars: 0
- Repository created: 2026-09-01
- Repository last pushed: 2026-10-21
- Publisher type: User
- Listing: https://agentpluginsdirectory.com/plugins/rolecasting--nisavid
- Schema: https://agent-plugins.org/schemas/1.0.0/plugin.schema.json

## What rolecasting does, in the publisher's words

Rolecasting is an Agent Plugins v1 package for choosing bounded worker topology and selecting an executable model and reasoning effort after delegation is settled. Claude Code uses a native manifest projected from the same canonical package identity.

Delegation freezes independent dimensions: target family, surface, version, and executor; child, peer, or external relationship; leader-owned or user-owned ownership; native-tool, task-api, CLI, app-server, or remote-API transport; and the observed and consumer-minimum product-attested, controller-observed, or self-reported assurance for each evidence dimension. Model choice remains separate. A transport never determines the relationship, and a user-owned peer requires explicit user consent.

Initial operational support uses separate native-child adapter profiles for ChatGPT Codex and Codex CLI/TUI. Both freeze the complete plan and assurance minima before native spawn and bind same-leader launch, status, and result observations afterward. A separate verification digest and strict Boolean preserve completed-but-blocked or unverified results as unusable.

From the project README, punctuation lightly normalized. Full text: https://raw.githubusercontent.com/nisavid/provingkit/HEAD/plugins/rolecasting/README.md

## Skills

- choosing-agent-models: Use when selecting a model or reasoning effort for an agent, subagent, Task, peer, or agent definition after delegation has been chosen.
- delegating-cross-agent-work: Use when deciding whether and how to delegate across child, peer, external, leader-owned, or user-owned worker surfaces.

Descriptions come from the frontmatter of each SKILL.md, punctuation lightly normalized.