skill-master

Use this agent to extract structured plugin/skill development intent from user descriptions or conversation context. Determines what capability is needed, component type, target audience, and checks for overlap with existing installed plugins. <example> Context: User describes a capability they want to build. user: "I want to build something that monitors my GitHub repos for stale PRs" assistant: "I'll use intent-distiller to clarify the capability scope and check for existing overlap." </example> <example> Context: User has a vague plugin idea. user: "I need a plugin for my daily workflow" assistant: "I'll use intent-distiller to extract structured intent from this request." </example> <example> Context: User wants to extend an existing plugin. user: "Add a skill that generates changelogs to my dev-workflow plugin" assistant: "I'll use intent-distiller to determine scope and check if this capability already exists." </example>

Use this agent to review Claude Code plugin artifacts (skills, agents, hooks, commands) from the AI executor perspective. Finds logic bugs, trigger mechanism issues, execution feasibility problems, and edge cases. Examples: <example> Context: User just created a new skill and agent. user: "Review the new design-drift skill and agents" assistant: "I'll use the plugin-reviewer agent to review these artifacts." </example> <example> Context: User wants to audit an entire plugin before release. user: "Audit the dev-workflow plugin" assistant: "I'll use the plugin-reviewer agent to review all skills and agents in dev-workflow." </example> <example> Context: A skill is behaving unexpectedly. user: "Review run-phase, something seems off in the review step" assistant: "I'll use the plugin-reviewer agent to check for logic issues." </example>

Check the YAML frontmatter and file organization.

Use this agent to scan ALL installed plugins for trigger overlap and conflict across plugin boundaries. Goes beyond single-plugin D5/D9 checks to detect cross-plugin routing ambiguity that causes skills from different plugins to compete for the same user input. <example> Context: After creating a new skill, verify it doesn't conflict with existing plugins. user: "Check if my new 'code-review' skill conflicts with anything installed" assistant: "I'll use trigger-arbiter to scan cross-plugin trigger conflicts." </example> <example> Context: User notices unexpected skill routing. user: "When I say 'review', the wrong skill triggers" assistant: "I'll use trigger-arbiter to detect which skills compete for that trigger." </example> <example> Context: Pre-release audit of a new plugin. user: "Scan all trigger conflicts for the skill-master plugin before release" assistant: "I'll use trigger-arbiter to check each skill against all installed plugins." </example>

1. Read the `description` field of the skill being reviewed

Use when the user says 'master', 'build a plugin', 'create a skill', 'create an agent', 'review plugin', 'audit plugin', 'iterate skill', 'improve trigger', 'package plugin', or wants to create, review, iterate, or package Claude Code plugins and components. Single entry with 4 routes: - create: brainstorm → design → scaffold (plugin-dev) → eval baseline (skill-creator) → review → iterate - review: 9-dimension audit from AI executor perspective + cross-plugin trigger conflict detection - iterate: fix → re-eval (skill-creator) → compare baseline → verify - package: full plugin or single component into target project