Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By a-organvm
Takes a dense prompt, autonomously decomposes it into parallel dimensions, wraps each in a self-contained assignment envelope, dispatches to specialized subagents, and reconciles their returns into a unified output.
npx claudepluginhub a-organvm/a-i--skills --plugin coliseum-from-grainUse this agent to autonomously compose a single self-contained autonomous-work-assignment envelope from a dimension specification — when you have a named dimension and need a dispatch-ready envelope, when you want an independent envelope draft to cross-check the assignment-composition skill's output, or when you want the envelope built in isolation from the rest of the coliseum (useful when one dimension's envelope is failing the no-pingpong gate and needs a fresh attempt). Returns a 9-section envelope ready to be added to phase-2-assignments.md. <example> Context: A dimension was identified but the auto-composed envelope failed pingpong-detector gate. user: "Re-do assignment-003's envelope from scratch." assistant: "I'll dispatch the assignment-composer agent with the dimension spec and the failure notes." <commentary> Fresh composition without inheriting the failed envelope's framing. </commentary> </example> <example> Context: Mid-orchestration, one dimension turned out to need a different domain expert than originally planned. user: "We need a security-flavored version of assignment-002." assistant: "I'll use assignment-composer to draft a security-domain-tuned envelope for that dimension." </example>
Use this agent when a dense or minimal prompt has been received and you need an independent reading of what the prompt implies before composing any plan. Trigger when the user issues a "grain of sand" — a small surface that plainly encodes a large coliseum, or when you want a second-pass dimensional read to cross-check your own. The agent returns a structured dimensional analysis (verbatim grain, implied scope, forbidden moves, named constraints, candidate dimensions across six lenses, dispatch-budget recommendation) without itself doing dispatch. Examples: <example> Context: User issues a paragraph-long prompt that implies multiple parallel domains. user: "I want a plugin that handles autonomous work assignments — chunked, self-contained, no ping-pong" assistant: "Before I plan, let me get a second read on what this implies dimensionally." <commentary> A dense grain — second-pass dimensional read prevents premature collapse into one obvious axis. </commentary> </example> <example> Context: You have surfaced dimensions yourself and want adversarial cross-check. user: "Continue." assistant: "I'll invoke grain-reader to cross-check the dimension set I surfaced." <commentary> Using the agent as a second-perspective check on your own Phase-1 output. </commentary> </example>
Use this agent as the mandatory gate between Phase 2 (assignment composition) and Phase 3 (dispatch) of the coliseum-from-grain protocol. The agent reads phase-2-assignments.md and verifies each envelope is self-contained enough that a recipient subagent would not need to ping-pong back asking for clarification. Returns PASS / PARTIAL / FAIL per envelope, with specific quoted defects when fail. Trigger automatically from coliseum-dispatch skill before dispatch; trigger manually when you suspect an envelope is leaky and you want an adversarial check; trigger after any envelope is rewritten to re-verify. <example> Context: Phase 2 just completed; you are about to dispatch. user: "Proceed." assistant: "Per the protocol, I'll run pingpong-detector on phase-2-assignments.md before dispatch." <commentary> Mandatory gate; do not skip. </commentary> </example> <example> Context: User skeptical that an envelope is truly self-contained. user: "Is assignment-004 really dispatch-ready or are we kidding ourselves?" assistant: "I'll have the pingpong-detector audit that envelope specifically." </example>
Wraps each surfaced dimension as a self-contained 9-section autonomous-work-assignment envelope — scope, context, success criteria, allowed tools, return format, handoff — all the recipient subagent needs to execute without coming back. Use when invoked by coliseum-orchestrator as Phase 2; when dimensions are named and the next step is to make each independently dispatchable; or when the user asks "compose this as an assignment." The no-pingpong gate validates each envelope before dispatch.
Dispatches a composed set of assignment envelopes to domain-expert subagents in parallel, in a single message with multiple Agent tool calls. Enforces the no-pingpong gate via the pingpong-detector agent before any dispatch fires. Use when invoked by the coliseum-orchestrator as Phase 3; when envelopes are already composed and the next step is parallel execution; or when the user asks to "fan out" or "dispatch in parallel." Produces a dispatch log capturing what was sent, when, and where returns land.
Threads the four-phase grain-to-coliseum transformation — given a dense prompt (the "grain"), autonomously surfaces parallel dimensions, composes self-contained 9-section handoff envelopes, dispatches in parallel, and reconciles returns into a composed whole under a mechanical compression gate. Use when a small-surface prompt plainly implies large scope; when the user invokes "grain → coliseum," "autonomous work assignment," "no ping-pong," or "chunk and hand off." Declines protocol when the grain is not fan-out shaped (≥3 independent dimensions).
Merges the parallel returns from a dispatched coliseum into a single composed whole — alchemy, not concatenation. Resolves tensions between returns explicitly, surfaces any BLOCKED records as named gaps, ensures no return is silently discarded. Use when invoked by the coliseum-orchestrator as Phase 4; when all (or recorded) returns from a Phase-3 dispatch are in and the next step is composition; or when the user asks to "reconcile," "compose the returns," or "merge into a whole." Produces the final artifact — phase-4-composed-whole.md — that closes the grain-to-coliseum loop.
Surfaces the parallel domain dimensions implicit in a dense or minimal prompt. Use when a user prompt is small on the surface but plainly implies multiple independent domains needing different expertise; when explicitly invoked by the coliseum-orchestrator skill as Phase 1; or when the user asks "what dimensions does this prompt encode" or "what axes does this break into." Produces a named dimension set where each dimension is independently executable and not a paraphrase of another.
Uses power tools
Uses Bash, Write, or Edit tools
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Comprehensive startup business analysis with market sizing (TAM/SAM/SOM), financial modeling, team planning, and strategic research
Comprehensive .NET development skills for modern C#, ASP.NET, MAUI, Blazor, Aspire, EF Core, Native AOT, testing, security, performance optimization, CI/CD, and cloud-native applications
Semantic search for Claude Code conversations. Remember past discussions, decisions, and patterns.
Collection of example skills demonstrating various capabilities including skill creation, MCP building, visual design, algorithmic art, internal communications, web testing, artifact building, Slack GIFs, and theme styling
Substrate-agnostic methodology for guiding any structural overhaul through five gated phases — landscape discovery, taxonomy design, environment configuration, systemic ingestion, governance evolution. Each phase produces a named artifact; an integrity auditor enforces gate criteria between phases.
A composable skill framework for AI agent orchestration -- 101 production-ready skill modules spanning creative, technical, enterprise, and governance domains, organized into a federated registry with multi-agent runtime support.
Part of ORGAN-IV: Taxis -- the orchestration and governance layer of the ORGAN system.
a-i--skills is a structured repository of 101 AI agent skills -- self-contained instruction modules that teach large language models how to perform specialized tasks in a repeatable, composable way. Each skill is a directory containing a SKILL.md file with YAML frontmatter (metadata for discovery and activation) and Markdown content (the actual instructions an agent follows).
The repository serves three distinct functions:
Skill Library -- A browsable catalog of 159 skills across 12 categories, from algorithmic art generation to security threat modeling, each with standardized metadata, optional helper scripts, reference documentation, and asset templates.
Orchestration Infrastructure -- Python tooling for skill validation, registry generation, health checking, and multi-agent bundle distribution. A built-in MCP (Model Context Protocol) server enables runtime skill discovery and planning.
Federation Specification -- A published protocol that allows third-party skill repositories to be discovered, validated, and consumed by any compatible agent, enabling a decentralized ecosystem of interoperable skill providers.
The skills themselves range from beginner-level single-file instructions to advanced multi-file modules with executable scripts, OOXML schema references, and comprehensive troubleshooting guides. Four document-processing skills (DOCX, PDF, PPTX, XLSX) demonstrate production-grade complexity -- these are the same skills that power Claude's native document creation capabilities.
| Dimension | Value |
|---|---|
| Total skills | 101 (97 example + 4 document) |
| Skill categories | 12 |
| Multi-agent runtimes supported | 4 (Claude Code, Codex, Gemini CLI, Claude API) |
| Total files | ~3,745 |
| Repository size | ~5.2 MB |
| Federation schema version | 1.1 (stable) |
| Skill spec version | Current |
AI agents are increasingly capable of executing complex, multi-step tasks, but their effectiveness depends heavily on the quality of instruction they receive. A generic prompt produces generic output. A well-structured skill -- with domain-specific vocabulary, explicit constraints, worked examples, and validation criteria -- produces expert-level output repeatedly.
The challenge is organizational: how do you manage dozens or hundreds of such skills across multiple agent runtimes, ensure they remain valid as specifications evolve, and enable external contributors to build compatible skills without centralized coordination?
This repository answers that question with three architectural decisions: