agentsclimarketplace

Domain expert router

Skill wonsukchoi/domain-experts/skills/domain-expert-router

all human experts into AI agents

Install
npx -y skills add wonsukchoi/domain-experts --skill domain-expert-router

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 9 stars9 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use whenever the user asks you to act as, consult, or reason like a specific job role/domain expert (e.g. "review this like a CFO," "what would a UX designer say," "act as a contracts lawyer") and this repo's roles/ directory is available. Finds the best-matching role SKILL.md and adopts it, or honestly reports the gap if none exists.

SKILL.md

4.7 KB, as published. Nobody here has run it

Domain Expert Router

What this does

This repo (domain-experts) is a growing library of job-role SKILL.md files, each written so an agent can adopt that role's actual reasoning process — not a job-description summary. This skill is the dispatcher: given a user's request for a specific role/expert judgment, find the best-matching role in the library and load it, or say honestly that it's not covered yet.

Coverage is a small fraction of the ~1,016 O*NET occupations tracked in ROADMAP.md (see its **Progress:** line for the current count — check it rather than assuming a number, since this file isn't auto-regenerated) — a no-match result is the common case, not an edge case. Handle it as a first-class outcome, not an afterthought.

Procedure

  1. Identify the role being asked for, in plain terms, from the user's request (e.g. "review this contract like a lawyer" → role: contracts lawyer).

  2. Find the best match. Run:

    npx domain-experts match "<the role/task in plain language>" --json

    (or, if the CLI isn't available, grep roles/*/SKILL.md frontmatter description fields for the closest fit — same idea, no tooling required.)

  3. If confident: true:

    • Read the matched file in full (best.file from the JSON, or the path the CLI printed).
    • Adopt its reasoning process for the rest of this task: follow its "Decision framework," reflect its "Mental models & heuristics," and match its "Communication style" section.
    • Mention which role file you loaded, briefly (one line), so the user knows the source — don't silently role-play without attribution.
    • If the role's frontmatter has maturity: draft, treat its content as solid but not yet practitioner-verified — still use it, no need to caveat every sentence, but don't claim it's been reviewed by a real practitioner if asked.

  4. If confident: false (no strong match) — this is a hybrid response, not a dead end:

    • State the gap honestly first: "No role for <X> in the repo yet." If the match command returned low-confidence candidates, mention them as the closest available, but don't pretend they're a real fit if they aren't.
    • Point at how to close the gap: check ROADMAP.md for the closest O*NET occupation code (if one exists) and mention that a PR following TEMPLATE.md/CONTRIBUTING.md would add it.
    • Then offer to proceed anyway: ask, or if the user's intent is clearly "just answer me now," proceed to reason as that expert using general knowledge — but flag it clearly and specifically, e.g. "Improvising this from general knowledge, not from a reviewed repo role — treat with more skepticism than a drafted role." Do not present improvised reasoning with the same confidence framing as a loaded SKILL.md.
    • Never silently fall back to generic reasoning while implying it came from a vetted role file.

Why the honest-gap step matters

The whole point of this repo is that a loaded SKILL.md encodes actually-considered first-principles reasoning, not off-the-cuff generalist output. Silently improvising and presenting it with the same weight defeats that purpose and erodes trust in the roles that are drafted. Saying "this isn't covered yet" costs nothing and is more useful than a confident-sounding guess mislabeled as expert judgment.

Example

User: "Review this vendor contract like our in-house counsel would."

  1. Role identified: contracts/commercial lawyer.
  2. npx domain-experts match "in-house counsel contract review" --json → confident match: lawyer-contracts.
  3. Read roles/lawyer-contracts/SKILL.md in full, adopt its worst-case-first reading heuristic, its risk-allocation decision framework, its US-commercial-law scope note.
  4. Respond: "Using this repo's lawyer-contracts role (scoped to US general commercial law) — here's the review: ..."

User: "What would a wildlife veterinarian say about this animal's symptoms?"

  1. Role identified: wildlife veterinarian — not in roles/.
  2. Match returns no confident hit.
  3. Respond: "No wildlife-veterinarian role in the repo yet (closest tracked-but-undrafted O*NET occupation would be 29-1131.00 Veterinarians — check ROADMAP.md). Want me to reason through this from general knowledge anyway? I'd flag that answer as unreviewed/improvised, not sourced from a vetted role."

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.