LFP

LazyCodex extension plugin that brings the full OMO feature set into Codex.

LazyCodex is the OMO Light edition for Codex. LFP keeps LazyCodex/OMO as the owner of agent behavior and adds the operational layer for model/provider setup, category-based model routing guidance, fallback guidance for saved override configurations, and model benchmarking.

LFP runs npx lazycodex-ai@latest install first, then registers this plugin in Codex, optionally configures a generic OpenAI-compatible provider only after operator consent, and writes saved model-routing choices to ~/.codex/lfp.json. It applies only Codex-supported primary model fields (model, model_reasoning_effort, service_tier) to existing LazyCodex/OMO agent TOMLs plus Codex global defaults when that mode is enabled.

Repository and LFP-owned issues live at https://github.com/islee23520/lazycodex-flavour-pack. If a failure is caused by upstream LazyCodex/OMO behavior rather than this flavour pack, register that issue on the upstream LazyCodex tracker instead.

lfp setup runs npx lazycodex-ai@latest install before applying LFP. It validates the configured agentsDir before writing its own install files and auto-generates model recommendations from provider /v1/models when the active provider can be queried.

Local development uses the checkout directly: npm run setup runs the LazyCodex preinstall step, then installs files from this working tree into CODEX_HOME/local-marketplaces/islee23520/plugins/lfp. Use node scripts/cli.mjs setup --skip-lazycodex-install only when intentionally testing local files without reinstalling LazyCodex first.

OpenAI-compatible provider setup is consent-gated. In interactive setup, LFP asks before writing a model provider into Codex config and records the answer under CODEX_HOME/.ledger/lfp/. Non-interactive setup skips provider installation unless consent has already been recorded.

Contents

• scripts/sync-agent-overrides-hook.mjs: quietly applies configured model overrides at session start and before prompt guidance.
• src/model/sync-agent-overrides.ts: reapplies supported three-primary-field model settings for existing LazyCodex/OMO agent targets from the configured override source.
• agent-configs/omo-agent-model-overrides.toml: packaged default model recommendation seed used before a user ~/.codex/lfp.json exists.
• agent-configs/codex-openai-compat-provider.toml: durable provider source for Codex OpenAI-compatible setup.
• agent-overrides/omo.json: legacy JSON override source retained for compatibility with older script callers.

Commands

npx @islee23520/lfp@latest setup
npx @islee23520/lfp@latest setup --no-tui
npx @islee23520/lfp@latest dry-setup
npx @islee23520/lfp@latest delete
npx @islee23520/lfp@latest doctor
npx @islee23520/lfp@latest sync
npx @islee23520/lfp@latest agent-config

npm test
npm run setup
npm run dry-setup
npm run delete
npm run doctor
npm run sync
npm run agent-config
npm run global-cli
npm run smoke:isolated

setup installs/enables LFP under CODEX_HOME/local-marketplaces/islee23520/plugins/lfp, removes obsolete LFP-owned helper agent TOMLs from older installs, and applies configured model-field overrides to existing LazyCodex/OMO agent TOMLs. Agent TOML sync is limited to the three primary model fields: model, model_reasoning_effort, and service_tier; global default sync remains limited to the same three fields for top-level config.toml and ulw.config.toml.

The canonical user config is ${CODEX_HOME}/lfp.json (normally ~/.codex/lfp.json) with schemaVersion: 2. It stores source, overrides, and rolePolicies in one JSON document. Setup and agent-config create or update this file; benchmark-models --apply writes winning model fields back to this same canonical path.

Fallback model guidance is available for saved override configurations that include fallback fields. Installed Codex agent TOMLs receive primary model fields only.

Interactive terminals get a Clack setup shell with confirm/cancel framing around the same setup work. Non-interactive setup, dry-setup, and doctor keep line-output behavior. Use setup --no-tui to force the legacy line-output setup path in a TTY.

Interactive setup enters a single model selection flow by default, covering Default Codex, ULW, and configured LazyCodex/OMO agents. Each model, reasoning-effort, and service-tier prompt shows what the setting affects, the current OMO/LazyCodex value, the vanilla LazyCodex recommendation where available, the LFP recommendation where one exists, and a short English guide for the agent role, tuning goal, and minimum capability. TUI choices mark (current) and (vanilla LazyCodex default) directly in the option label. Pressing Enter keeps and re-applies the shown value while still allowing edits; choosing Back to previous setting revisits the previous field or setup section. Line mode accepts b, back, :back, or previous for the same back-navigation flow. When provider models are discoverable, setup builds recommendations from the active provider inventory automatically. Saved choices are written into ${CODEX_HOME}/lfp.json before setup applies the overrides. Use --skip-model-prompt for non-guided setup.

When interactive model setup changes override values, LFP saves a schema-versioned JSON user copy at ${CODEX_HOME}/lfp.json. On later interactive setup runs after an npx/package patch, LFP asks whether you want to adjust model overrides; answering no keeps the saved settings without rerunning the per-agent prompts. Answering yes loads the saved copy and continues into the model selection flow. Older ${CODEX_HOME}/lfp/omo-agent-model-overrides.json, ${CODEX_HOME}/lfp/omo-agent-model-overrides.toml, and ${CODEX_HOME}/.ledger/lfp/omo-agent-model-overrides.toml copies are migrated automatically into the canonical JSON config path.

sync applies ${CODEX_HOME}/lfp.json directly to ${CODEX_HOME}/agents/*.toml without prompting. agent-config runs the same LFP model selector without reinstalling the plugin. It lists already-configured override targets and writes selections to ${CODEX_HOME}/lfp.json. Agent TOML writes are restricted to the three primary model fields on existing LazyCodex/OMO targets.

npm run global-cli builds the local checkout and installs it as the global lfp command, so newly built commands such as lfp sync are available through /opt/homebrew/bin/lfp or the active npm global bin.

Role recommendation policy defaults live in agent-configs/lfp-role-policies.toml. The canonical override location is ${CODEX_HOME}/lfp.json under rolePolicies; when present, those values take priority. The legacy sidecar ${CODEX_HOME}/lfp/lfp-role-policies.toml with [policies.<role>] sections still works for now. Model preference order stays code-managed.

dry-setup previews pending writes. delete removes the installed LFP plugin runtime and LFP marketplace/plugin config tables; it preserves LazyCodex/OMO state, provider config, and ${CODEX_HOME}/lfp.json. doctor reports plugin install state, upstream LazyCodex/OMO readiness, provider status, and pending override work.

smoke:isolated runs setup, saved user override restore, override sync, doctor, and Codex Apps cache cleanup against a temporary CODEX_HOME; it does not touch the real Codex install.

LFP prompt guidance scripts stay lightweight. The override sync path is the only path that mutates agent TOMLs, applying the configured three primary agent model fields; fallback prompt guidance remains read-only and points agents at the saved fallback configuration.

The packaged override configs resolve ${CODEX_HOME} at runtime, so the same release works across different user home directories and custom Codex homes without editing the shipped files.

Publish

GitHub Actions publish automation lives at .github/workflows/publish.yml. It runs when a GitHub release published event fires and can also be started manually with workflow_dispatch.

The workflow verifies the package with npm test and npm pack --dry-run, then publishes with npm publish --provenance --access public.

Required npm publishing setup

Configure npm trusted publishing for @islee23520/lfp against this GitHub Actions workflow. Token-based publishing is intentionally not used, so the workflow can publish with provenance without an OTP prompt.

If using npm Trusted Publishing instead of a token, configure npm for:

• Package: @islee23520/lfp
• Repository: islee23520/lazycodex-flavour-pack
• Workflow file: publish.yml

npm trust github @islee23520/lfp --repo islee23520/lazycodex-flavour-pack --file publish.yml

After auth is configured, publish by creating a GitHub release or manually running the workflow.