{
  "markdown": "# rapp-skill\n\n**Hand this package to Claude Code, GitHub Copilot CLI, Codex, or any agent harness,\nand it can run your whole RAPP ecosystem and move capabilities between RAPP agents\nand Agent Skills without a rewrite.**\n\nInstall a local agent brainstem with no API keys, keep it healthy, install and test\nsingle-file agents from two public catalogs, talk to it, and promote what works to\nAzure and Copilot Studio.\n\nThe package contains two discoverable skills:\n\n| Skill | Engine | Purpose |\n|---|---|---|\n| `rapp` | `skills/rapp/scripts/rapp.py` | Install, operate, test, and promote the RAPP ecosystem. |\n| `rapp-agent-converter` | `skills/rapp-agent-converter/scripts/toast.py` | Convert `agent.py` ↔ Agent Skill with a byte-exact linked-agent pair. |\n\n```bash\npython3 skills/rapp/scripts/rapp.py doctor\n```\n\n---\n\n## What this is\n\nRAPP is a three-tier platform: a **brainstem** on your laptop, **Azure Functions** in\nyour subscription, and **Copilot Studio** in front of your users. The same single `.py`\nagent file runs at every tier, so there is no rewrite between prototype and production.\n\nThis repository is the **skill that ties it together**. It reimplements nothing — it\ncalls the same installer one-liner a human would run and reads the same public catalogs\na human would browse. What it adds is a single, consistent command surface an AI can\ndrive without improvising.\n\n| Tier | Name | Where | What it gives you |\n|------|------|-------|-------------------|\n| 1 | Brainstem | Your laptop | The agent loop and memory on disk at `~/.brainstem`. Reached through the GitHub Copilot CLI, so no API keys to provision. |\n| 2 | Spinal Cord | Azure Functions | Always-on, cloud storage. Same agent file. |\n| 3 | Nervous System | Copilot Studio / Teams | Governed, published, in front of real users. |\n\n## Install the skill\n\n```bash\n# Claude Code\n/plugin marketplace add kody-w/rapp-skill\n/plugin install rapp\n\n# Any agent directory (Codex, OpenClaw, .agents)\ngit clone https://github.com/kody-w/rapp-skill ~/.agents/skills/rapp-skill\n```\n\nThen just ask: *\"install my brainstem\"*, *\"is my brainstem healthy?\"*, *\"find me an\nagent that tracks projects\"*, *\"test this agent\"*.\n\n## RAPP runtime commands\n\nRuntime operations route through `skills/rapp/scripts/rapp.py`, so an AI never\nhas to hand-write a `curl` call against the brainstem or a multipart upload.\n\n| Command | What it does |\n|---------|--------------|\n| `doctor` | Health check across every tier, with a fix line for anything broken. `--deep` adds a live chat round-trip; `--postmortem` reads what the last run did. |\n| `install` | Runs **the canonical installer one-liner**. `--dry-run` shows it without executing. |\n| `up` / `down` / `status` | Start, stop, and check the local brainstem. |\n| `search <query>` | Search RAR — 200+ single-file agents, ranked by AI critic score, votes and downloads. |\n| `agents list/install/remove/export` | Manage what is loaded. Installs are **SHA-256 verified** and hot-loaded without a restart. |\n| `store list/install` | RAPPstore — converged **rapplications** and **senses**. |\n| `chat \"…\"` | Send a turn and see the tool trace. |\n| `test <file>` | Load an agent through the brainstem's own loader and **actually run it** — sandboxed, timed out, contract-checked. |\n| `map` | Which repo owns which layer of the ecosystem. |\n| `tiers` | Where this machine stands across all three tiers. |\n| `memory` | Inspect and back up what the brainstem remembers. |\n\nAdd `--json` to any `rapp.py` command for structured output.\n\n## Agent Skill conversion commands\n\nConversion is a separate CLI owned by the sibling skill:\n\n```bash\npython3 skills/rapp-agent-converter/scripts/toast.py \\\n  convert path/to/foo_agent.py --to skill -o out/SKILL.md\npython3 skills/rapp-agent-converter/scripts/toast.py \\\n  convert out/SKILL.md --to agent\npython3 skills/rapp-agent-converter/scripts/toast.py \\\n  roundtrip path/to/foo_agent.py\n```\n\n`toast.py` does not accept `--json`. Its stdout is already a stable\nhuman-readable verdict and `inspect` emits JSON.\n\n## Zero to a working agent\n\n```bash\nrapp doctor          # what is missing\nrapp install         # the canonical one-liner\nrapp up              # start it\nrapp doctor --deep   # prove the model answers\nrapp search notes    # find something useful\nrapp agents install @kody-w/context_memory_agent\nrapp chat \"what tools do you have?\"\n```\n\nThe last command should list the agent you just installed.\n\n## Prerequisites\n\nPython 3.11+, `git`, `gh`, and a GitHub account (`gh auth login`). Nothing else — the\nruntime reaches models through the GitHub Copilot CLI, which is the preferred backend,\nso there are no API keys to provision. Whatever Copilot access your account has is\nbetween you and GitHub; this skill makes no claim about it.\n\nTiers 2 and 3 use your own Azure and Microsoft 365 subscriptions and need the `az`,\n`func`, and `pac` CLIs. `rapp doctor` reports which are present without assuming any.\n\n## Design rules\n\n- **Reimplements nothing.** The installer, the catalogs, and the brainstem are upstream.\n  The submitted `rapp-agent-converter` skill owns conversion. This package is the seam\n  between them; `rapp.py` does not grow a second converter.\n- **Integrity is not optional.** Every install is SHA-256 verified against the catalog.\n  A mismatch stops the install.\n- **Third-party code is sandboxed.** `test` runs agents in a subprocess, in a throwaway\n  working directory, with a timeout.\n- **Never diagnose from memory.** `doctor` reads the live system and prints a fix line;\n  the AI relays it rather than improvising shell surgery.\n- **No entitlement claims.** The tool is named; the subscription is not characterized.\n  Enforced by a test and by CI.\n\n## Repository layout\n\n```\nskills/rapp/\n  SKILL.md               the contract an AI reads\n  scripts/rapp.py        the engine — every command\n  references/            tiers, agent contract, troubleshooting\nskills/rapp-agent-converter/\n  SKILL.md               conversion operating procedure\n  scripts/toast.py       byte-exact agent.py ↔ Agent Skill engine\n  references/            RAPP contract + rapp/1 wire\n  assets/                executable sample cartridge\ntests/                   offline contract tests\n.claude-plugin/          Claude Code plugin + marketplace\n.codex-plugin/           Codex plugin\n.grok-plugin/            Grok plugin + marketplace\n.agents/plugins/         generic agent-directory marketplace\ngemini-extension.json    Gemini extension\n```\n\n## Tests\n\n```bash\npython -m pytest tests/ -q\n```\n\nThe suite is offline by design: no brainstem and no network required, so CI stays\nhonest on a machine with neither.\n\nThe converter's own selftest and sample roundtrip run in the same suite.\n\n## The ecosystem\n\n[rapp-installer](https://github.com/kody-w/rapp-installer) ·\n[rapp-brainstem](https://github.com/kody-w/rapp-brainstem) ·\n[RAR](https://github.com/kody-w/RAR) ·\n[RAPP_Store](https://github.com/kody-w/RAPP_Store) ·\n[rapp-twin](https://github.com/kody-w/rapp-twin) ·\n[rapp-map](https://github.com/kody-w/rapp-map)\n\nRun `rapp map` for the full picture.\n\n---\n\nMIT licensed. RAPP and the RAPP family of names are trademarks of the RAPP project.\nMicrosoft, Azure, Microsoft 365, Copilot, Copilot Studio and Teams are trademarks of\nMicrosoft Corporation; GitHub and GitHub Copilot are trademarks of GitHub, Inc. Named\nhere only to describe interoperability.\n",
  "bytes": 7375,
  "sha": "3f11e57ebe703e81d08ffd003b3eee9658b1bd3650782c60cafaf8a0c83a9bca",
  "repo_slug": "kody-w/rapp-skill",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_kody_w_rapp_skill_20304e8b/readme"
}