Design and architecture
The agent picks ituipath-safe-refactor-plan
Plan a behavior-preserving UiPath refactor after an architecture direction is selected. Use when a proven hotspot needs small implementation steps, safety nets, compatibility rules, stop conditions, and rollback points across workflows, packages, queues, selectors, or orchestration artifacts.
Install this skill
Claude Code
npx skills add 1aifanatic/uipath-boost --skill uipath-safe-refactor-plan --agent claude-code --global --yesCodex
npx skills add 1aifanatic/uipath-boost --skill uipath-safe-refactor-plan --agent codex --global --yesMake the structural change easy, then make the easy change, while keeping the automation demonstrably working after every meaningful step.
Maturity: core.
Ownership Boundary
This custom skill owns: Make the structural change easy, then make the easy change, while keeping the automation demonstrably working after every meaningful step.
Keep current product commands, schemas, artifact validation, live tenant operations, and policy administration with official UiPath skills.
Compose With Official UiPath Skills
Use official skills for current product commands and artifact contracts:
uipath-reviewuipath-testuipath-rpauipath-agentsuipath-maestro-flowuipath-maestro-bpmnuipath-maestro-caseuipath-api-workflowuipath-solution
Workflow
1. Establish current behavior
Require a selected direction from an architecture proposal, review finding, incident decision, or user choice. Read the PDD or SDD, acceptance tests, operational evidence, callers, public contracts, ADRs, and current validation results. State what must not change.
Completion criterion: The behavioral invariant and refactor boundary are explicit.
2. Verify the problem and alternatives
Inspect the relevant artifacts, confirm the cited friction, and compare deletion, extraction, consolidation, replacement, or no-change options.
Completion criterion: The selected direction solves observed pain better than its alternatives.
3. Build the safety net
Identify or add behavior-level tests, validation snapshots, sample transactions, comparison harnesses, and rollback artifacts at the highest correct seams.
Completion criterion: A structural mistake will produce a detectable failure.
4. Design the target seam
Use uipath-component-design where a public contract changes. Record hard-to-reverse choices and compatibility strategy.
Completion criterion: The target shape and transition contract are stable enough to sequence.
5. Sequence tiny safe steps
Plan prefactoring, expand, migrate, verify, and contract steps. Each step leaves the project valid or names an integration branch and final green gate when independent green is impossible.
Completion criterion: Every step has a check, rollback point, and bounded blast radius.
6. Define stop conditions
State thresholds that trigger rollback, re-planning, or escalation, including performance, business output, exception rate, and test regressions.
Completion criterion: The team knows when not to continue.
7. Publish the plan
Create dependency-ordered work items and route implementation to the owning official skills. Do not perform the refactor in this planning skill.
Completion criterion: An implementer can execute without inventing transition strategy.
Output Contract
- Behavioral invariants.
- Verified refactor rationale and alternatives.
- Safety-net plan.
- Tiny dependency-ordered steps.
- Rollback and stop conditions.
Guardrails
- Do not combine business behavior changes with structural cleanup unless separately approved.
- Do not replace a working pattern merely because another pattern is fashionable.
- Do not use file paths as the only durable description of intent.
- Do not execute source changes during planning.
- Route broad architecture surveying to
uipath-architecture-deepening; this skill sequences an already selected direction.
Example Requests
- "Plan extraction of duplicated broker rules into a library."
- "Refactor a 3,000-activity workflow safely."
- "Migrate a shared queue schema without stopping production."
Finish
Report completed work, observed evidence, the next official owner, and every blocker. Mark unobserved actions as pending.