Use Claude Code

/modernize is the main entry point. Each phase is also its own command, so you can run one step at a time or re-run a step after a change.

How /modernize works

graph LR
    A[Collect] --> B[Triage]
    B --> C[Analyze]
    C --> D[Assign]
    D --> E[Reality Check]
    E -->|approve| F[Schema Design]
    F --> G[Synthesis]

/modernize runs in one of three modes: chat, ui or both (the default). In ui and both it starts the local API and UI before the pipeline. The chat always shows the key numbers for each phase, so you can follow the run and decide from the chat alone. In both mode it also points to the matching UI view: the Sankey and assignment page at the Reality Check approval, the schema designs after phase 6, and the full report at the end. If the UI can't start (a busy port, a failed build, or no Node), the run switches to chat and continues.

Use /modernize docs/examples/wordpress/wordpress-collection.json --mode chat for a terminal-only run: no browser, no Node required, works in headless environments.

The individual commands

Command	What it does
/collect	Parses collector output and initializes a job
/triage	Detects workload signals and selects candidate engines
/analyze	Runs analysis agents for every triage-selected engine
/assign	Resolves query-to-engine assignments
/reality-check	Consolidates under-committed engines and validates with a model
/design-schema	Designs target schemas per engine (needs a model)
/synthesize	Produces the final report: decision report, engineering report, executive PDF/deck

Each command is defined in .claude/commands/ and reads the job state from .modernizer-state.json, so you can run them out of order to re-do a single phase after changing an earlier decision.

Analyzing your own database

Run the read-only collection script for your database engine instead of the sample zip:

# PostgreSQL (needs the pg_stat_statements extension)
psql -U <user> -h <host> -d <database> -t -A -f scripts/collect-postgresql.sql > my-collection.json

# MySQL (needs performance_schema)
mysql -N -u <user> -p -h <host> -D <database> < scripts/collect-mysql.sql > my-collection.json

The scripts only run SELECT on information_schema, pg_stat_statements or performance_schema. They never change your database. Then run /modernize (or /collect) and point it at the generated file instead of the sample.

Headless runs

claude -p "/modernize <collection.json> --auto --mode <chat|ui|both>" runs the whole pipeline without prompts. /modernize uses --mode when given and both otherwise, and --auto approves the one decision for you. CI uses this to test the pipeline against a real model; see ci/e2e-llm.sh and AGENTS.md for the full rules. Most users don't need it.