AI Pod OS

AI Pod Operating System — the discovery-to-specification half of a software delivery pod, packaged as agent skills.

Each skill is a role with a defined scope: which artifacts it owns, which tools it calls, and where it sits in the pipeline. A supervisor orchestrates the rest, so a vision can be decomposed into a backlog, prototyped, architected and estimated without re-explaining the process every time.

Install it once and the same skills are available in Claude Code, Codex, Cursor, Gemini CLI and Kiro.

Install

Claude Code

/plugin marketplace add IntelliasLabs/ai-pod-os
/plugin install ai-pod-os@ai-pod-os

Codex

npx codex-marketplace add IntelliasLabs/ai-pod-os --plugin --project

Gemini CLI

gemini extensions install https://github.com/IntelliasLabs/ai-pod-os

Cursor — install from Customize → Plugins.

Kiro — no plugin system; copy the skills into a directory Kiro scans:

./scripts/install-kiro.sh            # or: pwsh ./scripts/install-kiro.ps1

Per-host detail, the Kiro project-scoped install, and release steps are in docs/plugin.md.

Prerequisite

These skills read and write a shared backlog through MCP tools (workitem_*, backlog_*) and supervisor delegates through agents.invoke.*. The plugin ships skills only — connect the AI Pod backlog MCP server through your host's normal MCP configuration, or the skills will describe the process correctly but their tool calls will fail. See Prerequisite.

The skills

explore runs first and assembles context; supervisor orchestrates the rest.

Skill	Role
explore	Assembles a Change Context summary from existing specs, ADRs, NFRs and constraints
supervisor	Orchestrates the specialists and synthesizes their output
product-manager	Decomposes a vision into epics and features; owns prioritization and traceability
requirements-analyst	Turns product intent into SRS-grade requirements with testable acceptance criteria
designer	One self-contained interactive HTML prototype per user-facing feature
architect	Solution architecture, ADRs, NFRs, quality attribute scenarios, mermaid diagrams
estimation-agent	PERT three-point estimates rolled up with Monte Carlo simulation
presenter	Rewrites a finished, estimated spec into client-facing proposal prose

This is the specification pod: it takes a vision as far as an architected, estimated backlog and a client-facing proposal. Downstream delivery roles — implementation, testing, review, validation, security audit, release — are not part of it.

Full role descriptions and the common pipelines are in AGENTS.md, which doubles as the context file for Gemini CLI.

Repository layout

skills/                          the payload — <name>/SKILL.md plus references
AGENTS.md                        skill index and pipeline guide; Gemini CLI context file
.claude-plugin/                  Claude Code plugin + marketplace manifests
.codex-plugin/                   Codex plugin manifest
.agents/plugins/                 Codex marketplace manifest
.cursor-plugin/                  Cursor plugin + marketplace manifests
gemini-extension.json            Gemini CLI extension manifest
scripts/install-kiro.{sh,ps1}    Kiro installer (Kiro has no plugin system)
scripts/validate-plugin.test.mjs manifest and skill front matter validation
docs/plugin.md                   per-host install and release guide

The repository root is the plugin — every manifest points at ./, and all five hosts serve the same skills/ directory.

Development

node --test scripts/validate-plugin.test.mjs

The manifests must agree on name, version and description; every declared path must exist; every skills/<name>/SKILL.md must have front matter whose name matches its directory; and AGENTS.md must index every skill. CI runs this on every push, plus claude plugin validate . --strict.

Adding a skill is one directory under skills/ and one row in AGENTS.md — no manifest changes.

License

Apache-2.0. See LICENSE.