Proposes OpenAgent, a spec for defining AI agent identity (face, voice, writing style) in a single signed YAML file, enabling portability across different harnesses.
the agent was never the model or the harness. it's the identity. and right now that identity is scattered across a dozen files. proposal: OpenAgent. tiny file spec pins how an agent looks, sounds, and writes, and keeps it the same everywhere. switch harness - the character stays. ed25519 signed, so anyone can verify who authored it, offline, no server. remix lineage if you fork a persona. npx, MIT, zero install. here's the whole file: ```yaml openagent: "0.2" # spec version id: nova # lowercase-kebab handle — your stable id name: Nova role: Research Analyst org: # optional — your company/team name: Acme Labs face: ref: # public portrait URL — the card's hero anchor: "calm analyst, soft studio light, head-and-shoulders" sprite: # optional sprite sheet of expressions, for animation/feed voice: audio: provider: google-gemini # or elevenlabs, openai … (vendor-neutral) base: Kore # voice name within that provider style: "calm, precise; states the finding, then the source" written: rules: # how the agent writes - "Lead with the finding." - "Cite it or it didn't happen." sample: "Three sources agree; the fourth's a 2019 outlier." behavior: "I dig up primary sources, never guess, and cite everything." posts_about: [research, sourcing, evals] provenance (a did:key signature) is added automatically when you mint a card ``` try it, 10 seconds: already running an agent? try it now: install the openagent skill (npx skills add 5dive-ai/skills --skill openagent) and make your openagent card and show it to me. it reads its own role, voice, and behavior, emits a valid persona.yaml, renders its rarity card, and can open a PR into the open registry (if you want public to see it). quick tip: if you want your agent to have a specific face - send it the photo first (just say "use this as your face") before it generates anything. your agent's identity should be a file you sign and own. it's early, v0.2. rather it get torn apart now than later. what would you actually need in a persona spec to use it? open to all new ideas and spec improvements.
The article proposes that organizations adopting AI coding agents should create a company-wide AGENTS.md file, similar to a human onboarding doc, to standardize agent behavior and context.
The author argues that an AI agent is best understood as a folder of markdown files containing business knowledge and instructions, separate from the model and harness, enabling portability between rapidly improving harnesses.
The article explores the emerging need for identity verification and permission management for AI agents, as agent-to-agent workflows and autonomous systems become more common, proposing concepts like signed tool manifests and agent certificates.
The article argues that as AI agents autonomously perform actions in shared workspaces, clear attribution of each action to both the agent and the accountable human is necessary for oversight and trust. Without proper identity and audit trails, teams cannot safely delegate more complex tasks to agents.
A guide to writing an effective SOUL.md file for AI agents, covering eight essential sections that define identity, voice, and boundaries to compound agent performance.