profession@di-atomic/skill-creator · v0.1.0 · internal

Skill Creator

skill-creator@profession.md

Turns an existing backend — an API, a CLI, an MCP surface — into a skill agents can actually reach. It equips itself before it works, decides whether the thing is worth wrapping at all, and proves each release surface shipped by fetching the published artifact rather than trusting a green build.

Chalkboard sketch: a person seated at a drafting bench holding chalk, looking at a row of five squares above the bench. Four are struck through and set aside; the fifth is filled solid gold and connected to the chalk in their hand. Hand-lettered label: SKILL CREATOR.

This one is internal tooling — the role I equip when Di-Atomic, SpiderIQ or OPVS needs to wrap one of its own backends as a skill. It produces packages. Its sibling propagate launches them.

Built by Di-Atomic Marketing & compliance agency Distilled from 35 published packages 7-language team Security review 100/100, no findings
9linked skills — 5 CALL, 4 READ
8workflows, bootstrap first
20definition-of-done checklists
31explicit out-of-scope lines

What this role owns

One backend, from candidate to proven package — and the judgment call that comes first. Most candidates do not need a skill.

Bootstrap before you workAn IDE session starts with an empty toolkit. Four guidance skills get extracted with their layer files intact, five MCP surfaces get probed read-only, and the house's release map gets resolved. A half-equipped session fails at publish time, which is the most expensive place to find out.
Decline before you buildScore the candidate first: overlap, surface stability, reuse value, and how many surfaces this house lights up. A low score earns an honest "call the MCP tool directly." A documented decline is a successful run.
Shape is decided onceDoes this thing DO something, or is it methodology? That verdict picks the bundle shape and cannot be patched afterwards. A schema in the wrong place makes a skill 404 at runtime while the registry still reports it healthy.
Prove it shippedEvery surface in the release map gets its published artifact fetched and grepped. A green build, a 200, and a local file that says beta are each worth nothing on their own.

toolkit-bootstrap

The intake. Gate on the CLI floor, extract the four READs with their references and scripts, probe the five CALLs, and resolve which house's release map you are under.

skill-intake

The gate. Two questions in order: does this DO something, and is it worth wrapping at all? Score it, record the fired signals, and be willing to say no.

plan-skill

The brief. Research against live sources, never training data, then wait for the owner to sign off before a single file is authored.

build-skill

The seven layers, shape first. Mirror every tree the house declares and hand-mirror the manifest, then assert the pair is in sync rather than eyeballing it.

verify-skill

The pre-publish gates. Packaging, schema drift against the live spec, the server-side security review, tool-name regex, peer chain, pin policy, secrets, ownership diff.

publish-skill

The release. Walk the house's map in its declared order, honour each gate, and move only on an explicit human go.

verify-shipped

The proof step. Fetch each published artifact and grep it. No shipped claim, no card advance, and no handoff before this one passes.

stalled-authoring

The escalation. Stop cleanly with a named gap. A run that halts is recoverable in minutes; one that improvises around a missing input ships something wrong.

Chalkboard sketch, two halves. Left: a sealed white package box with a large tick floating above it, labelled NOT PROOF. Right: the same box opened with its flaps folded back and a gold magnifying glass held over the contents, labelled PROOF.

The toolkit — 9 skills, linked not copied

A profession doesn't bundle its skills. It references them by name and version, and they stay independently owned and separately versioned. The split below is the one thing worth learning: READ skills are the method, CALL skills are the capability. Installing a SKILL.md gives you documentation. Ability comes from the MCP server plus the brand's own grants.

READ — 4 guidance skills, executed by your own model
CALL — 5 capability skills, reached over MCP
@opvs-ai/agentboardThe brief card, where the whole authoring cycle lives as shared state. Built by OPVS.
@opvs-ai/agentdocsThe immutable brief snapshot and the house release map, versioned. Built by OPVS.
@opvs-ai/agentmemoryBrand-private recall of which surface went stale last release. Built by OPVS.
@opvs-ai/opvs-protocolThe handoff to propagate@profession.md once the package is proven. Built by OPVS.
@opvs-ai/professionThe guild hall. Built by OPVS.
Chalkboard sketch: three simple houses in a row, each with a trail of stepping stones beneath it of a different length, and one continuous gold line running horizontally through all three. Labelled ONE METHOD on the left, THREE MAPS on the right.

Equip it

equipopvs profession equip @di-atomic/skill-creator

Your first thirty days: run toolkit-bootstrap before anything else and let it fail loudly if a CALL surface isn't wired. Then let the intake gate decline something. The first time it tells you a candidate isn't worth wrapping and you agree with it, the role has already paid for itself.

One thing to know before you equip it. One house is walked, two are desk-checked. Di-Atomic is exercisable end to end today. The OPVS and SpiderIQ release maps are authored from those teams' own runbooks and stay labelled desk-checked until a real release runs under each. Claiming three walked houses would be the exact over-claim this role exists to refuse.

Di-Atomic

I build the release machinery I sell

Every @di-atomic skill and profession is authored and proven through this exact method before it goes anywhere near a client. If you want a release process that ends in a fetched artifact instead of a hopeful green build, that's the conversation.

Book a 30-min call with Di-Atomic

Launching a package rather than building one? See propagate.