If the user's intent does not match this skill, activate /plugin-creator:plugin-lifecycle to route it.

Plan Plugin Refactoring

<plugin_name> $ARGUMENTS </plugin_name>

Execute the phases in order. Before Phase 1, create separate tracking tasks named for assessment, design, task-file creation, plan-index update, context gathering, and final handoff. Mark each in progress before execution and complete only after verifying its output. After each phase, display its key findings and output paths before continuing.

Phase 1: Assess

Dispatch plugin-creator:plugin-assessor with:

• Input: ./plugins/<plugin_name>/ and the request to assess structural correctness, quality issues, and refactoring opportunities.
• Context: identify the plugin manifest and any skill, command, agent, hook, MCP, and reference content under the plugin root.
• Execution ownership: the agent runs its complete assessment protocol. Do not restate its phases, schemas, or scoring rules in the dispatch prompt.
• Required output: the standard assessment produced through its preloaded /plugin-creator:assessment-reporting skill, including marketplace readiness, scored findings, affected paths, severities, orphan classifications, and recommended execution roles.
• Completion: STATUS: DONE plus the inline report or report path. STATUS: BLOCKED must name the missing input.

Verify the report exists, contains the score and marketplace-readiness result, and accounts for every discovered component. Then activate /plugin-creator:audit-skill-lifecycle <plugin_name> and /plugin-creator:audit-agent-lifecycle <plugin_name> to add semantic lifecycle findings.

Activate /plugin-creator:audit-skill-completeness for individual marketplace candidates, low-scoring skills, or when the user requests deep quality evaluation. Otherwise record that this optional audit was not run.

Load /plugin-creator:assessment-reporting and apply its Finding Admission section before deciding which Phase 1 findings enter the executable plan. Preserve every finding in the reports.

Phase 1 is complete when the structural report and both lifecycle audit results are present, every optional-audit decision is recorded, and each finding is marked admitted or advisory.

Phase 2: Design

Dispatch plugin-creator:refactor-planner with:

• Input: the complete Phase 1 assessment and audit results.
• Target: .plugin-creator/plans/refactor-design-{plugin-slug}.md, where the slug is the plugin directory name.
• Context: plugin root, current component paths, all findings, and validator output. Treat SK006/SK007 findings as the source of truth for complexity; do not copy numeric thresholds.
• Required content: every admitted finding mapped to a concrete transformation, target path, dependencies, preserved behavior, and safe parallelization group. Keep advisory findings in the report with their rationale, but do not create a transformation. Include split plans, agent optimizations, documentation changes, and orphan resolutions only when present in admitted findings.
• Completion: write the target and return STATUS: DONE with its path, or STATUS: BLOCKED with the missing input.

Read the design and verify every Phase 1 finding has one disposition. Phase 2 is complete when the design exists and has no undisposed finding.

Phase 3: Plan Tasks

Dispatch plugin-creator:refactor-planner with:

• Inputs: .plugin-creator/plans/refactor-design-{plugin-slug}.md and the existing .plugin-creator/plans/REFACTOR-PLAN.md when present.
• Targets: .plugin-creator/plans/tasks-refactor-{plugin-slug}.md and the matching plan-index entry.
• Task contract: create tasks only for admitted transformations. Each task names its ID, type, target, dependencies, priority, execution role, at least three acceptance criteria, required inputs, expected outputs, at least three verification steps, and safe parallel peers. Express ordering through dependencies rather than temporal estimates.
• Routing: SKILL_SPLIT to /plugin-creator:refactor-skill; AGENT_OPTIMIZE to plugin-creator:subagent-refactorer; DOC_IMPROVE to plugin-creator:ai-doc-optimizer; read-only audits to plugin-creator:skill-auditor; upstream synchronization to plugin-creator:skill-content-updater; description-only work to /plugin-creator:write-frontmatter-description.
• Verification tasks: validate the completed plugin, then update plugin documentation when the design requires it.
• Completion: write both targets and return STATUS: DONE with paths and parallelization groups, or STATUS: BLOCKED with the missing input.

Verify both files and confirm dependency edges match the design. Phase 3 is complete when every designed transformation has an executable task and every task has a verification path.

Phase 4: Gather Context

Dispatch plugin-creator:refactor-planner with:

• Inputs: the design, task file, and ./plugins/<plugin_name>/.
• Write target: the task file's Context Manifest section.
• Required content: current summaries for each target, existing patterns to preserve, cross-references, and external dependencies. Pass source paths; do not substitute summaries for unread source.
• Completion: return STATUS: DONE with the updated task-file path and manifest counts, or STATUS: BLOCKED with the missing input.

Verify the Context Manifest exists and accounts for every task target. Phase 4 is complete when no task lacks source context or dependency evidence.

Final Handoff

Verify every tracking task is complete, then return:

STATUS: DONE
Plugin: <plugin_name>
Assessment: {inline or report path}
Design: .plugin-creator/plans/refactor-design-{plugin-slug}.md
Tasks: .plugin-creator/plans/tasks-refactor-{plugin-slug}.md
Plan index: .plugin-creator/plans/REFACTOR-PLAN.md
Context manifest: present
Parallel groups: {groups or "none"}
Next action: /plugin-creator:implement-refactor {plugin-slug}

If execution cannot continue:

STATUS: BLOCKED
Phase: {phase}
Reason: {specific missing input or failed verification}
Completed: {verified outputs}
Remaining: {uncompleted outputs}