Agent Skills Matrix

This table reflects intended skill usage for .claude/agents/*.md.

Quick read:

  • Language-heavy agents use language/domain skills on demand.
  • Docs/content/EM agents stay intentionally narrow.
  • High-impact workflows are manual-invocation skills.
Agent Allowed skills
codebase dotnet, python, typescript, flutter, go, java-spring, node-express, react-next, ux-responsive, ruby-rails, rust, sql-migrations
orchestrator dotnet, python, typescript, flutter, go, java-spring, node-express, react-next, ux-responsive, ruby-rails, rust, sql-migrations, project-bootstrap, docs-validation, agent-diagnostics, blogger, brutal-critic
planner dotnet, python, typescript, flutter, go, java-spring, node-express, react-next, ux-responsive, ruby-rails, rust, sql-migrations, project-bootstrap, docs-validation, agent-diagnostics
review dotnet, python, typescript, flutter, go, java-spring, node-express, react-next, ux-responsive, ruby-rails, rust, sql-migrations, docs-validation, agent-diagnostics
docs docs-validation, project-bootstrap, agent-diagnostics
em-advisor project-bootstrap, agent-diagnostics, docs-validation
blogger blogger, brutal-critic
brutal-critic brutal-critic, blogger

Notes

  • Skills load on demand to reduce context overhead.
  • Keep skill usage narrow by role to reduce tool/context noise.
  • Avoid preloading many skills in subagents unless explicitly needed.

Skill Trigger Guide (Canonical)

Use this as the canonical trigger reference for when agents should load skills.

Task intent Skill to load Typical agents
Language/framework implementation or review Matching stack skill (dotnet, python, typescript, flutter, go, java-spring, node-express, react-next, ruby-rails, rust, sql-migrations) codebase, review, planner, orchestrator
Responsive UX across phone/tablet/desktop ux-responsive codebase, review, planner, orchestrator
Docs quality checks (lint/link/structure) docs-validation docs, review, orchestrator
Agent/package diagnostics and configuration checks agent-diagnostics docs, review, planner, orchestrator, em-advisor
Project bootstrap scaffolding project-bootstrap docs, planner, orchestrator, em-advisor
Content drafting blogger blogger, orchestrator
Content critique and scoring brutal-critic brutal-critic, blogger, orchestrator

Trigger policy:

  • Load skills only on demand for the active task/phase.
  • Start with one relevant skill; add a second only for explicit cross-domain needs.
  • If stack/domain is ambiguous, ask a clarifying question before loading.

Skill Scope Policy (Current)

  • Keep the current core-only skill set.
  • Expand only when demand, gap, ownership, and licensing checks are all satisfied.

Command Skill Routing (Pattern C)

Command skills fork context and delegate directly to a target agent. They are invoked by the user via /skill-name [argument] and never load into the calling agent’s context.

Command skill Target agent Argument hint Purpose
/api-docs docs [module, file, or endpoint path] Generate API reference documentation
/architecture-decision docs [decision topic or system name] Create an ADR
/architecture-review review [system, component, or design document] Review design for quality attributes
/code-review review [file, PR, or scope — blank for current changes] Security, performance, and style review
/content-review brutal-critic [content text, file path, or topic] Score and critique content quality
/create-readme docs [project name or path] Generate or improve a README
/generate-tests codebase [file, class, or function name] Generate targeted tests
/1-on-1-prep em-advisor [person] [context] Prepare for a focused 1-on-1
/plan-project orchestrator [feature, objective, or epic] Multi-phase implementation plan
/refactor-plan planner [target module, file, or scope] Safe staged refactor strategy
/security-audit review [scope, file, component, or 'full project'] Security-focused code and config review
/blog-post blogger [topic or title] Draft a blog post

Note: Command skills never expose themselves in the agent tool list. They bypass the calling agent entirely — the delegated agent receives only the skill body + $ARGUMENTS as its prompt.