From Skills
Author a new subagent (.md) that matches this project's house agent structure — frontmatter with routing guards, Your job, Operating principles, Workflows, Working-inside-a-workstream, and What-you-don't-do — kept self-contained (bundled refs, no external URLs). Use when creating a new role/subagent, adding to the agent fleet, or turning a role description or a Competency Dossier into an agent. Do NOT use to edit skills (see write-a-skill) or to run the agent once built.
How this skill is triggered — by the user, by Claude, or both
Slash command
/skills:write-an-agentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Create a new agent file that matches the house template, so the whole fleet stays consistent.
Create a new agent file that matches the house template, so the whole fleet stays consistent.
Write it to the plugin's agents/<name>.md (when extending this plugin) or the project's
.claude/agents/<name>.md.
@researcher
(role / job-requirement research) to produce a Competency Dossier and persist it to
research/<role-slug>.md (project: .claude/research/…; bundled: ${CLAUDE_PLUGIN_ROOT}/research/…),
named by discipline. Author the agent from it — its Recommendation for the AI version maps straight
onto the agent's job, tools, and scope.templates/agent.md and fill every section
(delete any you deliberately omit — leave no placeholder behind).name — lowercase-kebab, unique, matches the filename. This becomes the @role and subagent type.description — what routes here plus explicit guards (a "Do NOT use…" clause). The orchestrator
sees only this when routing, so make it discriminative against the other agents.model — choose deliberately: inherit to follow the session, or pin (claude-opus-4-8,
claude-sonnet-4-6, claude-haiku-4-5-…) for cost/latency.tools — least privilege. Read/Grep/Glob to analyse; add Write/Edit only if it produces files;
Bash/WebSearch/WebFetch only if needed. Omit Agent unless the role legitimately orchestrates.## Your job; ## Operating principles that open with the disposition (what the role cares about / is allergic to) then the
imperative rules — only persona-heavy roles need a separate ## Identity; for deep roles an
## Expertise that synthesises the dossier (load-bearing facts inline) and references
research/<role-slug>.md by path — never a re-paste; inline a ## Workflows entry only if the
procedure is worth spelling out; ## What you don't do scoped by capability/domain, handing
out-of-scope work back to the orchestrator — don't hardcode peer agent names (routing is
description-driven; the fleet varies per project).PROTOCOL.md (it travels with the run).name lowercase-kebab, matches filenamedescription has a "Use when…" trigger AND a "Do NOT use…" guardtools least-privilege; model chosen on purpose<placeholders>PROTOCOL.mdExpertise synthesises the dossier and links research/<role-slug>.md — not a re-pasteWhat you don't do is capability-scoped and hands back to the orchestrator — no hardcoded peer namesProvides CDSS development patterns for drug interaction checking, dose validation, clinical scoring (NEWS2, qSOFA), and alert classification integrated into EMR workflows.
npx claudepluginhub lucasvscn/skills --plugin skills