Commands and Workflow Reference

This page is a quick guide to LingXi's current foreground capabilities and lower-level runtime commands.

Foreground Capabilities

LingXi currently keeps its explicit workflow surface focused on two capabilities:

Capability	Purpose	Typical output
task	Turn a rough request into an engineer-ready task document	.lingxi/tasks/*.md
vet	Challenge task quality before implementation starts	VetReport (structured output)

These capabilities run only when explicitly invoked.

Background Capabilities

LingXi also runs a background memory loop:

Capability	Purpose	Typical output
session-distill	Distill durable engineering taste from historical sessions	writes to .lingxi/memory/ and updates state and index
memory-retrieve	Retrieve the smallest useful set of memory for current work	memory brief / retrieval hits
memory-write	Govern and persist adjudicated candidates	notes + INDEX.md updates

Common Scripts

These are the most useful runtime scripts in the current LingXi surface:

lx-bootstrap

npm run lx:bootstrap
# or
node scripts/lx-bootstrap.mjs

Purpose:

• initialize .lingxi/, write AGENTS.md
• create or merge .codex/config.toml, .codex/hooks.json (Codex adapter)
• generate .codex/agents/lingxi-session-distill.toml
• create or merge .claude/settings.json (Claude Code adapter)
• generate .claude/agents/lingxi-session-distill.md, copy skills to .claude/skills/
• write CLAUDE.md, generate .lingxi/setup/automation.session-distill.toml
• register the session-distill automation (Codex only)

lx-distill-sessions

npm run lx:distill-sessions
# or
node scripts/lx-distill-sessions.mjs

Purpose:

• scan session artifacts
• select valid sessions
• hand selected sessions to the single-session worker

Generic Conversation Memory Consumption

The current mainline no longer ships a supported manual lx-memory-brief command.

For meaningful repository turns, LingXi injects the smallest relevant memory brief automatically through a repo-local UserPromptSubmit hook (Codex or Claude Code).

Notes:

• task and vet still call the underlying retrieval flow directly rather than going through the hook
• on native Windows, setup still generates hook config, but the current Codex runtime does not execute hooks natively yet

lingxi-setup

npm run lx:setup
# or
node scripts/lingxi-setup.mjs

Purpose:

• initialize the LingXi runtime directory layout .lingxi/
• write AGENTS.md only when missing
• create or merge .codex/config.toml, .codex/hooks.json (Codex adapter)
• generate .codex/agents/lingxi-session-distill.toml
• create or merge .claude/settings.json (Claude Code adapter)
• generate .claude/agents/lingxi-session-distill.md, copy skills to .claude/skills/
• write CLAUDE.md only when missing
• generate .lingxi/setup/automation.session-distill.toml

Supports --host parameter: codex, claude, all (default).

lx-create-automation

npm run lx:create-automation
# or
node scripts/lx-create-automation.mjs

Purpose:

• register or update LingXi's session-distill automation from setup artifacts

Uninstall

npm run lx:uninstall
# or
yarn lx:uninstall

For non-interactive environments:

yarn lx:uninstall --yes

Testing And Release Checks

The main repository currently uses these checks to keep installation, workflow, and memory behavior aligned:

• npm test
• install-manifest-exists
• install-manifest-coverage
• install-manifest-version-sync

Next

• Core Workflow
• Memory System
• FAQ