riel-claude-code-refactoring-workflow

name

riel-claude-code-refactoring-workflow

description

Use when planning, running, monitoring, or troubleshooting the packaged rielflow divide-and-conquer refactoring workflow. Applies to `claude-code-refactoring-divide-and-conquer`, duplicate-scavenge refactoring, concurrent slice review via `claude-code-refactoring-slice-review`, generated refactoring implementation plans, implementation/review loop handling, blocked verification, dependency-owned agent mocks, and final plan archival.

Riel Claude Code Refactoring Workflow

Use this skill when the user asks to refactor rielflow through the divide-and-conquer workflow, asks whether the refactoring workflow is progressing, or asks to create/continue a maintainability refactor using workflow review loops.

Workflow Bundle

• Parent workflow: claude-code-refactoring-divide-and-conquer
• Parent package id: claude-code-refactoring-divide-and-conquer
• Child workflow: claude-code-refactoring-slice-review
• Child package id: claude-code-refactoring-slice-review
• Typical generated plan path: impl-plans/active/refactoring-runtime-boundaries.md
• Completed plans must move to impl-plans/completed/ and be indexed in impl-plans/README.md.

Standard Run

Run from the repository root:

rielflow workflow package checkout claude-code-refactoring-divide-and-conquer
rielflow workflow package checkout claude-code-refactoring-slice-review
rielflow workflow run claude-code-refactoring-divide-and-conquer \
  --variables '{"workflowInput":{"requestedOutcome":"Improve rielflow maintainability using divide-and-conquer refactoring.","targetPaths":["src/workflow","src/cli","src/lib.ts","packages",".rielflow/workflows"],"excludePaths":["dist","node_modules","/tmp","impl-plans/completed"],"maxSlices":6,"constraints":["Do not stage, commit, or push unless the user explicitly asks.","Do not revert unrelated dirty worktree changes.","Keep each implementation pass to one bounded task.","Update any generated refactoring plan progress as tasks are implemented."],"verificationPreferences":["Run narrow Bun tests for touched code first.","Run workflow validate for changed workflow bundles.","Run git diff --check before completion."]}}' \
  --output json --verbose --no-auto-improve

Adjust requestedOutcome, targetPaths, excludePaths, and constraints to the user's request. Keep the constraints explicit so worker nodes preserve unrelated worktree changes and do not stage/commit/push during refactoring.

Duplicate-Scavenge Mode

Use duplicate-scavenge mode when the user asks to find duplicated implementations, parallel custom implementations of the same concept, or code that should be unified behind one helper, API, or abstraction.

Keep this as a mode of claude-code-refactoring-divide-and-conquer; do not create or run a separate duplicate-only workflow. The workflow prompts own the detailed runtime contracts:

• Step 1 slices explicit targetPaths first, then broad package/processing groups when the request is not targeted.
• claude-code-refactoring-slice-review owns per-slice duplicate evidence.
• Step 3 owns merged duplicateGroups and implementation task contracts.
• Steps 4 through 6 implement and review only the Step 3-authorized task fields.

Example input:

{
  "workflowInput": {
    "requestedOutcome": "Run duplicate-scavenge refactoring on the selected paths and consolidate safe duplicate candidates.",
    "refactoringMode": "duplicate-scavenge",
    "targetPaths": ["src", "packages", ".rielflow/workflows"],
    "excludePaths": ["dist", "node_modules", "impl-plans/completed"],
    "maxSlices": 8,
    "constraints": [
      "Do not stage, commit, or push unless the user explicitly asks.",
      "Do not revert unrelated dirty worktree changes.",
      "Preserve public behavior and public APIs unless the plan explicitly authorizes a change.",
      "Prefer consolidation only when counterpart implementations have matching semantics or documented differences."
    ],
    "verificationPreferences": [
      "Run focused tests for each consolidated behavior.",
      "Run workflow validate when workflow bundles change.",
      "Run git diff --check before completion."
    ]
  }
}

When reviewing results, reject broad abstractions that only remove superficial similarity. Accepted duplicate-scavenge tasks should preserve the Step 3 contract fields: owned paths, counterpart paths, behavior to preserve, known differences not to collapse, consolidation target, conflicts, and narrow verification commands.

Monitoring

Record the parent session id from workflow output. Monitor with:

rielflow session status <session-id> --output json

Useful compact status query:

rielflow session status <session-id> --output json \
  | jq '{status, queue, nodeExecutionCounter, currentStepId, counts:.nodeExecutionCounts, lastTransition:(.transitions[-1] // null), lastExec:(.nodeExecutions[-1] // null)}'

Implementation nodes can be quiet for several minutes while an agent backend edits and verifies. Treat silence as a stall only after checking the process/session state.

Expected Loop

The parent workflow should:

1. Have the manager normalize scope and constraints.
2. Slice the codebase by package or related processing group.
3. Fan out concurrent child claude-code-refactoring-slice-review runs.
4. Merge findings into an implementation plan.
5. Implement exactly one bounded task.
6. Self-review.
7. Post-review.
8. Loop to implementation when needs_revision or plan_remaining is true.
9. Stop at workflow-output when complete or when only accepted/blocked residual work remains.

Do not manually edit files while an implementation node is actively running unless the workflow has failed or the user explicitly redirects the task.

Verification

After workflow completion, independently verify instead of relying only on workflow-recorded evidence:

bun run typecheck:server
bun run lint:biome
git diff --check

Run focused tests for touched areas. Common examples:

bun test src/workflow/adapters/claude-code.test.ts src/workflow/adapters/claude.test.ts
bun test src/workflow/runtime-execution-contracts.test.ts src/workflow/engine-fanout.test.ts src/workflow/adapter.test.ts
bun test src/workflow/save.test.ts src/workflow/session-store.test.ts src/workflow/manager-session-store.test.ts
bun test src/package-boundaries.test.ts

Validate changed workflows:

rielflow workflow validate claude-code-refactoring-divide-and-conquer --output json
rielflow workflow validate claude-code-refactoring-slice-review --output json

If mock scenarios changed, run them with workflow run --mock-scenario.

Agent Package Mock Rule

Do not create rielflow-local replacements for dependency agent mocks. rielflow adapter tests should consume dependency-owned test SDK exports, such as:

• claude-code-agent/sdk/testing
• claude-code-agent/sdk/testing

If those exports are missing, update or fix the owning dependency package (codex-agent, claude-code-agent, or cursor-cli-agent) rather than teaching rielflow dependency internals.

Completion

When the plan is complete:

• Mark the plan status and exit criteria complete.
• Move the plan from impl-plans/active/ to impl-plans/completed/.
• Update impl-plans/README.md.
• Report completed tasks, residual risks, verification results, and whether anything was staged/committed/pushed.

Only commit or push when the user explicitly asks.