AI Workflow Cleanup Sprint

What shipped

• Runnable local service-delivery console at http://localhost:3000.
• Dependency-free Node HTTP server with static frontend and JSON persistence.
• Three seeded demo workspaces for a home-services company, marketing agency, and professional-services office.
• Tool register with add, edit, delete, CSV import, deterministic scoring, decisions, notes, and evidence.
• Workflow inventory with test notes, failure modes, review gates, rollback steps, and scoring.
• Ranked cleanup queue combining tools and workflows.
• Cleanup action board for keep, pause, remove, replace, and repair recommendations.
• Workflow repair worksheet with before state, repair steps, after state, measurement, owner, review gate, and rollback.
• Export generators for CSV, JSON, executive report, risk register, cleanup plan, owner follow-ups, workflow repair summary, and operating checklist.
• Smoke test covering the core sprint loop.
• Supporting service assets in docs/: delivery checklist, intake questions, and sales one-pager.

Architecture

• Used built-in Node modules instead of Express to avoid dependency installation and keep the sandbox easy to inspect.
• Kept business logic in src/scoring.js, src/importers.js, src/exporters.js, src/templates.js, and src/store.js so a later SaaS or CLI could reuse it.
• Stored each workspace as one JSON file under data/workspaces/ for transparent local persistence.
• Used deterministic scoring only. No Claude, Anthropic, OpenRouter, LLM API, OAuth, or external system access is required.
• Built a dense operational UI with tables, score chips, action lanes, and export controls rather than a landing page.

Trimmed scope

• No authentication, multi-tenant SaaS, billing, deployment, or production database.
• No live integrations with Google Workspace, Microsoft 365, Zapier, Make, n8n, CRM, accounting, helpdesk, or SaaS management tools.
• No PDF export. Markdown and CSV are the primary client-ready artifacts.
• No automated subscription cancellation or destructive workflow changes.
• No AI-generated summaries. Reports are produced from structured local data.

Limitations

• CSV import supports common tool and expense columns, but workflow CSV import is provided only as a sample file in this pass.
• Scoring is useful for prioritization but not an authoritative risk, legal, privacy, compliance, or security assessment.
• Workspaces are local JSON files with no concurrent editing controls.
• UI validation is lightweight and assumes a trusted local consultant operator.
• Evidence attachments are text references only, not uploaded files.

Verification

Command run:

``bash npm test ``

Result: passed. The test created a workspace, imported sample tools, added workflows, generated scores, recorded a cleanup action, saved a repair worksheet, and wrote required exports under tmp/smoke-exports/.

Additional syntax checks passed for server.js, public/app.js, and scripts/smoke-test.js.

Local server start was attempted in this Codex sandbox, but socket binding was rejected by the environment with EPERM on both 0.0.0.0:3000 and 127.0.0.1:3000. The server is configured for local use with HOST=127.0.0.1 and should run normally in a standard developer shell.

Suggested next steps

• Add workflow CSV import and richer column mapping previews.
• Add a scoring rubric editor so Michael can tune factor weights per client type.
• Add PDF export from the Markdown deliverables.
• Add a monthly retainer review mode with changed-cost and new-tool tracking.
• Add optional adapters for client-provided exports from Zapier, Make, n8n, Microsoft 365, Google Workspace, or expense systems.
• Add a disabled-by-default OpenAI-compatible report drafting adapter after deterministic exports are stable.