MeowKit

CLI Reference

The mewkit command line, grouped by what you are trying to do.

MeowKit ships one CLI package, mewkit. Run it from your project root; there is nothing to install globally.

Find a command

You want toPage
Create a project, configure it, check it is healthy, upgrade itProject Setup
Write a plan, create a task, see where a task stands, record a supervised runPlanning and Tasks
Review a pull request across several sittingsReview
Read or write memory, check the context budget, use the wikiMemory and Knowledge
Export the kit to Codex or Cursor, or see what they supportPortability
See what a session cost, what it did, what the kit containsCost and Telemetry

Get started

# 1. Scaffold MeowKit into your project
cd your-project
npx mewkit init

# 2. Configure environment
npx mewkit setup

# 3. Verify installation
npx mewkit doctor

# 4. Create your first task
npx mewkit task new --type feature "Add user authentication"

Two things worth knowing up front

npx mewkit init both scaffolds and updates. The first run opens a toolkit picker (Claude Code, Codex, Cursor, with Claude Code preselected) and copies the full system into .claude/ for each provider you choose; --target <provider> skips the picker. Later runs are a smart update: unchanged core files are overwritten, files you edited are left alone, new files are added, and there is no picker.

Everything else is a runtime toolbox for daily use. None of it is required to run MeowKit, and none of it changes your project unless the command says it does.

Every flag

The table below is generated from the same catalogue that configures the parser and prints --help, so it cannot drift from what the CLI actually accepts. The family pages above carry the worked examples and the failure recovery; this is the lookup.

Generated from the CLI catalogue. Edit packages/mewkit/src/cli/command-catalogue.ts, not this table.

Every command

FlagWhat it does
--helpShow help
--versionPrint the CLI version

mewkit init

Scaffold or update MeowKit in the current project.

FlagWhat it does
--betaUse the beta release channel
--dry-runPrint what would happen without writing
--forceOverwrite an existing .claude/
--target <value>Skip the toolkit picker and create one toolkit
--profile <value>Install a subset: core, developer, product, atlassian, security, research, full
--migrateAfter unpack, prompt for export targets
--migrate-globalUse global install paths instead of project-local
--skill-packs <value>Comma-separated skill packs to install
--mcp-profiles <value>Cursor only: MCP profile names, or 'all', to merge into .cursor/mcp.json
--allow-cloud-mcpCursor only: second opt-in for applying an MCP profile to a repo with a git remote

mewkit upgrade

Upgrade MeowKit to the latest version.

FlagWhat it does
--yesAnswer prompts affirmatively
--betaUse the beta release channel
--checkReport without changing anything
--listList available versions
--no-cleanupKeep the downloaded release directory

mewkit validate

Validate .claude/ project structure.

FlagWhat it does
--target <value>Validate a generated provider target instead of a .claude/ project
--mode <value>Force authoring or flat-copy mode instead of auto-detecting
--portableInclude portable provider contract checks
--strictTreat WARN as failure (exit 1)
--workflowRun only the workflow.yaml drift check
--gatesRun only the gate-authority contract check
--ownershipRun only the artifact ownership-completeness check
--agentsRun only declared agent-contract conformance checks
--packsRun only the pack-manifest coherence and safety check
--rulesRun only the routing-table-breadth check
--capabilitiesRun only the capability manifest checks
--substrateRun only the responsibility-substrate drift and untagged check

mewkit capabilities

Inspect or resolve the capability manifest.

Subcommands: list, explain, resolve, view, bootstrap, projections.

FlagWhat it does
--jsonEmit machine-readable JSON
--intent <value>Resolve this intent to the capability that owns it
--provider <value>Scope the projection to one provider
--writeWrite the generated view rather than printing it
--recordRecord the result to the task record; --no-record skips it

mewkit budget

View token usage and cost log.

Subcommands: context.

FlagWhat it does
--jsonEmit machine-readable JSON
--profile <value>Estimate one context profile instead of all of them
--session <value>Filter to the current session, or to a specific session id
--monthlyAggregate by month instead of the last ten entries
--day <value>Filter to today, or to a specific YYYY-MM-DD
--fail-over <value>Exit non-zero when a profile exceeds N tokens

mewkit memory

Manage agent memory (lessons and patterns).

Subcommands: capture, render-views, seed-from-md.

FlagWhat it does
--strictTreat WARN as failure (exit 1)
--clearClear all memory, with confirmation
--statsShow sessions captured and patterns learned

mewkit setup

Guided post-scaffold configuration.

FlagWhat it does
--only <value>Run a single setup step
--system-depsInstall system-level dependencies

mewkit doctor

Diagnose common environment issues.

Subcommands: provenance.

FlagWhat it does
--reportPrint a shareable diagnostic report
--providersInclude provider contract diagnostics
--stateInclude state taxonomy diagnostics
--explainExplain the provenance result rather than summarizing it
--consolidationShow the consolidation ledger; status is not a runtime-availability claim
--hard-gatesLive-probe the hard gates (plan, privacy, injection block)

mewkit status

Print version and config summary.

mewkit task-state

Durable task record.

Subcommands: show, update.

FlagWhat it does
--jsonEmit machine-readable JSON
--status <value>Task status: active, blocked, done
--step <value>What just finished
--next <value>The next action
--plan <value>Path to the plan this record joins
--blocker <value>Record a blocker, repeatable
--verification <value>Record a verification as ref=result, repeatable
--evidence-ref <value>Record an evidence reference, repeatable
--capability-decision <value>Record a capability decision as capId|decision|reason, repeatable

mewkit advice

Bounded supervision checkpoints for --advice; records evidence, approves nothing.

Subcommands: begin, commit, status, validate-packet.

FlagWhat it does
--jsonEmit machine-readable JSON
--provider <value>Runtime recorded on the receipt
--next <value>The next safe action recorded on the dossier and receipt
--run <value>Supervision run id (embedded supervision requires one)
--skill <value>The skill running the checkpoint, e.g. mk:fix
--stage <value>Checkpoint stage: GUIDE, RESCUE, REVIEW, RECHECK
--checkpoint <value>Named checkpoint id; repeating one is idempotent
--disposition <value>The returned disposition, recorded verbatim
--outcome <value>What the parent did with it: adopted, rejected, deferred
--reason <value>Why the outcome was chosen; required even when the directive was adopted
--question <value>The exact question the checkpoint asked
--directive <value>The returned directive, summarized
--correction <value>A required correction, repeatable
--evidence-pointer <value>An evidence pointer, repeatable
--slug <value>Slug used in the receipt filename; defaults to the run id
--task-id <value>Active durable task id, or omit for none
--evidence <value>Path to the workflow evidence index a correction supersedes, or the packet/brief to validate
--correction-kind <value>source (keeps Gate 1) or scope (invalidates it)
--packet-kind <value>validate-packet shape: input, output, or brief (direct consult)
--release-stage <value>Ship scope the call is budgeted against: prepare, release, publish

mewkit orient

Resume orientation from durable task state; reads records only.

FlagWhat it does
--jsonEmit machine-readable JSON
--recordRecord the result to the task record; --no-record skips it

mewkit context

Repo-context evidence.

Subcommands: resolve, check, record.

FlagWhat it does
--jsonEmit machine-readable JSON
--task <value>Task id to record the evidence envelope against
--root <value>Project root to resolve the path against

mewkit task

Create and list task files.

Subcommands: new, list.

FlagWhat it does
--activateActivate the task on creation; --no-activate opts out
--status <value>Filter the list by status: draft, in-progress, blocked, review, done
--type <value>Task type: feature, bug-fix, refactor, security
--priority <value>Task priority
--allInclude completed and backlog tasks

mewkit migrate

Export MeowKit to external coding-agent tools.

FlagWhat it does
--yesAnswer prompts affirmatively
--dry-runPrint what would happen without writing
--forceOverwrite user-edited targets on conflict
--only <value>Limit to item types: agents, commands, skills, config, rules, hooks
--providersInclude provider contract diagnostics
--allExport every item type rather than the default subset
--globalUse global install paths instead of project-local
--source <value>Migration source root
--source-version <value>Version to record for the migration source
--installInstall the runtime for the target provider
--reconcileReconcile an existing export instead of writing fresh
--skip-configDo not convert configuration
--skip-rulesDo not convert rules
--skip-hooksDo not convert hooks
--reinstall-empty-dirsRecreate directories that were emptied by a prior run
--respect-deletionsHonor deletions.json rather than reinstating removed files
--all-rulesMerge every rule into the provider instruction file
--include-mcpAlso convert .mcp.json servers into the provider MCP config
--include-unportableInstall Claude Code skills for a provider with no adapter (experimental)

mewkit providers

Show the effective provider support matrix and enforcement levels.

FlagWhat it does
--jsonEmit machine-readable JSON
--lifecycleShow the capability-adapter and lifecycle matrix

mewkit explain-support

Alias of providers.

FlagWhat it does
--jsonEmit machine-readable JSON
--lifecycleShow the capability-adapter and lifecycle matrix

mewkit visual-plan

Visual plan contracts.

Subcommands: validate, status, approve, rehash, export, prepare-feedback, apply-feedback, patch, edit, view.

FlagWhat it does
--jsonEmit machine-readable JSON
--forceOverwrite user-edited targets on conflict
--checkPre-apply gate for apply-feedback
--port <value>Bind to a fixed port instead of a random one
--openAuto-launch the browser
--no-openDo not launch the browser
--revision <value>Pin approve to the exact reviewed revision
--format <value>Export format
--ops <value>Operations file for prepare-feedback
--op <value>Operations file for patch
--batch <value>Feedback batch id for apply-feedback
--receipt <value>Record apply-feedback outcomes to this file

mewkit review

High-assurance PR review, resumable across sittings via a session id.

Subcommands: prepare, read, coverage, compose, submit, cleanup.

FlagWhat it does
--jsonEmit machine-readable JSON
--remote <value>Git remote to fetch the pull request from
--session <value>Review session id from review prepare
--as <value>Read the diff as one named persona
--replyPost the composed review back to the pull request
--confirm <value>Confirmation token required for the irreversible submit step

mewkit inventory

List harness artifacts with governance metadata.

FlagWhat it does
--jsonEmit machine-readable JSON
--checkFail if README or index counts drift from reality
--substratePrint the responsibility-by-coverage substrate matrix
--staleShow only deprecated or experimental artifacts
--criticalShow only criticality=critical artifacts
--portable-missingShow artifacts whose runtime is not portable
--emitWrite the substrate view rather than printing it
--emit-countsRewrite README and index count numbers to match reality

mewkit plan

Plan lifecycle.

Subcommands: status, check, approve, archive.

FlagWhat it does
--jsonEmit machine-readable JSON
--dry-runPrint what would happen without writing
--by <value>Approver name recorded on the approval receipt
--activateActivate the task on creation; --no-activate opts out

mewkit trace

On-demand trace recall.

Subcommands: score, audit, propose.

FlagWhat it does
--jsonEmit machine-readable JSON
--id <value>Trace record id
--friction <value>Record a friction note against the current task
--responsibility <value>Scope to one substrate responsibility
--commitPersist the proposal rather than previewing it

mewkit wiki

Long-term project knowledge.

Subcommands: context, init, propose, approve, search, reindex, handoff.

This command reads its own flags, which are owned by its subsystem rather than by the CLI catalogue.

mewkit index

Build or refresh the derived SQLite index over the append logs.

FlagWhat it does
--jsonEmit machine-readable JSON

mewkit query

Read-only aggregate queries over the derived index.

FlagWhat it does
--jsonEmit machine-readable JSON
--task <value>Task id to scope the query to
--presetsList the built-in query presets

mewkit pack

Manage install packs.

Subcommands: list, add, remove.

FlagWhat it does
--jsonEmit machine-readable JSON
--yesAnswer prompts affirmatively
--betaUse the beta release channel

mewkit verdict-gate

Check a review verdict against the active plan.

mewkit docs-manifest

Generate or drift-check the docs reference manifest.

FlagWhat it does
--jsonEmit machine-readable JSON
--checkFail when the committed manifest no longer matches the live .claude/ tree
--writeRegenerate the committed manifest

On this page