The AcademyAdvanced
Build Your Own Agents, Skills & Commands
What you'll build
Your own working AI-native primitive set, built on valuefirstagent.com: a real agent (front-matter-valid, with a wired expertise reference), at least one skill (valid YAML front matter, correct tier), and one command (authored to the ratified command template) — all runnable, and put through a real 'does it operate?' certification the same way this org certifies its own agents.
The live cadence
How the live thing runs for this path.
Office Hours as a builder clinic to start — author between sessions, bring your agent/skill/command to be reviewed live by the faculty who own this craft (Aegis for agent quality, Hone for skill/command architecture). Cohort later.
The build, step by step
What you'll do, in order.
- Study the living architecture as worked examplesBefore building, study real ones: open a few agent definitions in the full VFT agent roster, read a skill's front matter, read the command template. The org's own architecture IS the textbook. (open any file under agents/, the roster orientation in agents/README.md, the index in agents/manifest.json, and .claude/commands/_TEMPLATE.md.)
- Author a skill that passes the front-matter contractWrite your first skill with valid YAML front matter (title, tier, load_discipline, category, owner_agent, provenance) and the right tier semantics. Keep the two axes distinct: `tier` is arch (always-loaded) vs domain (demand-loaded); `load_discipline` further says when a domain skill loads — always, demand, or per-agent. The contract is enforced, so 'valid' is objective.
- Author a command to the ratified templateBuild a command following _TEMPLATE.md: Purpose, Execution Context, Inputs, Flow, Procedure, Outputs, Verification — surface-agnostic, inheriting global rules rather than restating them; the five typed calls used correctly.
- Build and ground your agentOn valuefirstagent.com, build an agent and ground it the way this org requires: an expertise reference that equips it to challenge and lead — not a yes-man. Learn the three-leg standard: externally grounded, challenge-and-lead, certified-against. (agents/RE-ONBOARDING.md Gate 0 + checklist A.)
- Certify that it operatesPut your primitive set through a real 'does it actually run?' certification — the constructive-evidence model where the agent does one real task as proof, exactly how this org certifies its own. (the /certify model + lint pack-wiring checks.)
- Bring it to the builder clinicJoin Office Hours and have your agent, skill, and command reviewed live by Aegis (agent quality + re-onboarding) and Hone (skill & command architecture). The live critique is where a working primitive becomes a trustworthy one.
Where the build becomes real
You build it on valuefirstagent.com.
the build destination: author and run your own agent there, with a skill and a command wired to it. The most literal build handoff of all six.
Join
Join the next live session for this path.
Tell us where to reach you and what you want to build. We will send you the next live session details.
Keep watching
Recent conversations to learn from.
Value-First AI DailyOctober 8, 2026
A Dot From Scratch
Chris Carolan starts an OpenAI dot on a ChatGPT account with no instructions or chat history. With only Google connected, it reads his work back to him, names…
Open the episodeThe Road to UNBOUNDOctober 7, 2026
The Road to UNBOUND - Season 2, First Stop: AI Taking the Joy Out of the Work: Passing Plates or Cooking, and Don't Delegate the Things You Love
Kyle Jepson gave the first stop of Season 2 the question the hour kept returning to: "…when we're working with AI, are we just passing plates or are we…
Open the episode
Value-First AI DailyOctober 6, 2026
YouSpot Writes Back
youspot's network email, its first built from Chris Carolan's inbox, drafted two reconnect notes. He judged them on air: one came close, one he would not send,…
Open the episodeThe next step
You're building. Here's where it goes next.
And then, the rest of the path: