/ripast

Use RipIDE for mechanical refactors across TS, JS, JSX, and Vue files. Run from the project root with Node 22.13+. Follow repository worktree and approval rules; keep unrelated work separate. Ripgrep is optional. If it is missing, Git discovers tracked and untracked files, including local edits. Use direct edits for small local changes, prose, strings, and comments. During architecture work, design contracts with direct edits. Use RipIDE for the supported mechanical steps. Svelte markup and arbitrary custom codemods are unsupported. For provider replacement, use --source-scope to keep imported aliases and public barrel names. Read replacement scope and limits before choosing direct edits for aliases or barrels.

Launcher

Use a supplied executable first, then an existing project CLI, then a global ripide. For repeated use, prefer installing ripide and ripide-vue globally once with the project manager. The package is ripide. Do not launch the retired @ripast/cli package. Unless an executable or version is supplied, check the installed and registry versions once per task. If the installed CLI is older, use the exact registry release through a temporary launcher. Keep project dependency pins unchanged. Announce any authorized global upgrade before installing it. Reuse the selected executable and version throughout the task. Do not reinstall before each command. Skip installation and version probes when a CLI is supplied. If setup is needed, read installation and launchers, including one-off and restricted environments.

Workflow

1. If a move crosses packages, first check destination dependencies and exports. Otherwise, run a known mutation without redundant discovery.
2. If uncertain, resolve the target and review a dry run first. Read command syntax and examples when needed.
3. Use --apply --profile agent; keep type verification enabled. Use --scope for a known declaration file.
4. If a check is supplied, append it with && in the mutation's shell call. Do not run it separately or invent checks.
5. Read the result. Inspect changed ranges only when the result or task leaves uncertainty. Stop after required checks pass.

Before manual edits, read the current source required by the editing tool.

Common commands, run only the requested operation and append its supplied check:

ripide rename useStore useAppStore --scope src/store.ts --apply --profile agent
ripide rename-file src/store.ts src/app-store.ts --apply --profile agent
ripide css-class-rename font-semibold font-medium --apply --profile agent
ripide css-custom-property-rename --apply --profile agent -- --foo-a --theme-b

Operation flags --artifact, --fields, and --minify require --json. --fields supports discovery and saved-evidence paging. The page command always emits JSON. For a complete plan and verification receipt, add --json --artifact <new-file.json> to the operation once. If discovery size is unknown, save a new artifact with the first command. For mutations, save an artifact when diagnostic or change details may need inspection. For JSON, read _tag for the outcome and data for command fields. Keep the original outcome, verification scope, stderr, and process exit. A page response cannot replace them. Keep complete artifacts outside model context. Inspect saved evidence only when the compact result leaves uncertainty. Agent pages target 4 KiB; stdout defaults to 32 KiB. Set --page-bytes and --max-bytes for your budget. Follow nextOffset; page sizes vary. Never advance by an assumed 40 results. For saved-evidence pages, read the view and continuation at data.view and data.view.nextOffset. Use ripide page --input <artifact.json> to inspect its root menu, then select the relevant JSON Pointer path. Do not assume the artifact root is an array. Command artifacts have different shapes. Use --fields when you know the required object fields. If a collection's item shape is unknown, inspect one item path before paging the whole collection. For nested file evidence, inspect /files/0, then select the required child path. Keep complete rows when their context helps. The page byte target is soft. Select nested paths when one item exceeds the response ceiling. Long source strings page into text parts. Read their source line numbers; offsets count parts. For known requests, batch NDJSON through page --session. Use live navigation only if your shell preserves stdin. Separate page calls are a cheap fallback. They reread evidence without scanning the project. Never repeat a mutation or project scan just to retrieve omitted evidence. If an applied mutation had no artifact, inspect current source and existing receipts. Do not reapply to create evidence. Read saved evidence for path selection and session requests.

Batch known commands. An exact CSS mapping needs no css-class-scan. CSS and template transforms have no type verification; run their project checks. For CSS custom properties, read supported cases and limits. Review source warnings and result limitations before apply. Value-only strings such as "--foo-a 1s" receive warnings. Inspect their CSS property before editing. Warnings include nested static candidates, ambiguous CSS values, and encoded style attributes. Prove their meaning before editing. Save complete warnings with --json --artifact <new-file.json>. Use the reference's agent review procedure for skipped sites. If no files changed, resolve the target, root, or configuration before retrying. Rename accepts two positional names. Use --scope <file> instead of a third positional path. Dry run is the default. There is no --dry-run flag. Quote one comma-separated --glob value. Do not repeat --glob or assume shell word splitting. Keep stderr and the exit code. Do not pipe command results through head or tail to judge success. If diagnostics increased, inspect the cause. Never disable verification to force success. Inspect more files only when results expose uncertainty. Repeat a successful check only after another edit or new failure. Search for one uncertainty that can change the next step. Keep full output in artifacts; read relevant evidence only. Read <launcher> <command> --help only when a needed flag is unclear.

For Nuxt auto-imports or verification scope and output limits, read verification and Nuxt.

For transient behaviour checks, use ripide check <export> --base <ref> with Vitest code on stdin. The CLI imports the function, test, expect, and vi. Add --from only for an ambiguous export. Read transient checks for mocks, integration obligations, and stale execution evidence.