# Lossless Codex Orchestrator

Give your main agent a memory and command layer for all your Codex projects and threads.

Codex is excellent at doing work. The hard part is managing all the work after
you have dozens or hundreds of threads across repos, customer projects, fixes,
reviews, and follow-ups.

LCO turns that scattered local Codex history into an operating layer your
Hermes agent, MCP client, OpenClaw agent, or custom agent can use. Your agent
can find the right project, understand what happened, see
what is blocked, prepare the next action, and keep moving without rereading huge
transcripts every time.

![Lossless Codex Orchestrator showing session cards, project memory, and agent command tools](https://raw.githubusercontent.com/100yenadmin/Lossless-Codex-Orchestrator-LCO/main/assets/readme/hero.png)

[![npm latest](https://img.shields.io/npm/v/lossless-codex-orchestrator/latest?label=npm%20latest)](https://www.npmjs.com/package/lossless-codex-orchestrator)
[![npm beta](https://img.shields.io/npm/v/lossless-codex-orchestrator/beta?label=npm%20beta)](https://www.npmjs.com/package/lossless-codex-orchestrator)
[![CI](https://github.com/100yenadmin/Lossless-Codex-Orchestrator-LCO/actions/workflows/ci.yml/badge.svg)](https://github.com/100yenadmin/Lossless-Codex-Orchestrator-LCO/actions/workflows/ci.yml)
[![CodeQL](https://github.com/100yenadmin/Lossless-Codex-Orchestrator-LCO/actions/workflows/codeql.yml/badge.svg)](https://github.com/100yenadmin/Lossless-Codex-Orchestrator-LCO/actions/workflows/codeql.yml)
[![License](https://img.shields.io/badge/license-PolyForm%20Noncommercial-blue)](LICENSE)
[![PRs welcome](https://img.shields.io/badge/PRs-welcome-brightgreen)](CONTRIBUTING.md)

```bash
npm install -g lossless-codex-orchestrator@latest
lco doctor
lco find "billing bridge"
```

If this helps your main agent stay on top of Codex work, a star helps other
agent builders find it. ⭐

[Setup](docs/SETUP.md) · [OpenClaw Plugin](docs/OPENCLAW_PLUGIN.md) · [Agent Skill](skills/lossless-openclaw-orchestrator/SKILL.md) · [Vision](VISION.md) · [Privacy](docs/PRIVACY.md) · [Hermes Boundary](docs/HERMES_ADAPTER_BOUNDARY.md) · [Contributing](CONTRIBUTING.md) · [AGENTS.md](AGENTS.md) · [Security](SECURITY.md) · [Code of Conduct](CODE_OF_CONDUCT.md) · [Release Notes](docs/releases/CHANGELOG.md) · [License](LICENSE)

## Why It Exists

When you use Codex heavily, the problem stops being "can an agent code?" and
becomes "can my main agent understand all the work already in motion?"

LCO gives that main agent a local memory and command surface:

| Without LCO | With LCO |
| --- | --- |
| Threads are scattered across many sessions and projects. | Your agent can search and triage them from one local index. |
| Every handoff starts with rediscovery. | Prepared cards show the objective, blocker, status, and next action. |
| Big transcripts eat context. | Summary leaves and bounded expansion let agents read the right slice. |
| OpenClaw or another orchestrator has to guess what Codex did. | MCP/OpenClaw tools expose Codex state directly to the orchestrator. |
| "Continue this work" is risky because the target may be unclear. | Dry-run command packets show the exact thread and action before anything runs. |

The goal is simple: your orchestration agent should manage Codex work the way a
good technical operator would. It should know the projects, the active threads,
the stale work, the blocked work, the finished work, and the right next move.

## What It Does ✨

LCO is more than a transcript index. It builds an agent-readable operating
picture over local Codex work.

**Project and thread memory**

- Indexes local Codex sessions into a local SQLite database.
- Uses field-weighted FTS5 search for session-card discovery over titles,
  summaries, proposed plans, final messages, touched files, tool metadata, and
  prepared safe text. For remembered content phrases, use `lco grep` or
  `lco expand-query`.
- Blends relevance, recency, identifier matching, and query fallback so agents
  can find "the billing bridge plan" or "the PR that touched auth" without the
  exact thread id.
- Detects Codex JSONL format drift in `lco doctor` so broken imports are visible
  instead of silently missing work.

**Prepared state for agents**

- Creates prepared cards for Codex threads: objective, blocker, lifecycle state,
  next action, freshness, confidence, and source refs.
- Builds a prepared inbox so your main agent can start from "what needs my
  attention?" instead of raw search.
- Tracks lifecycle states such as completed, waiting for approval, watching an
  external check, needs resume, dirty worktree handoff, ready for review,
  stale/partial, and unknown.
- Keeps completed work visible, so finished lanes can still be found and cited.

**Summary leaves and bounded expansion**

- Splits large sessions into source ranges and deterministic summary leaves.
- Creates leaf refs for user prompts, plans, final messages, closeouts, touched
  files, tool metadata, and compaction markers.
- Lets an agent expand a small 1k-token brief or a deeper 4k-token evidence
  bundle instead of loading an entire transcript.
- Reports omissions when a brief is intentionally smaller than the underlying
  session.

**Operating picture tools**

- `lco_recent_sessions` shows recent or active Codex work as compact cards.
- `lco_attention_inbox` lists threads that need action, review, approval, watch,
  or blocker triage.
- `lco_project_digest` creates a project-level handoff brief from Codex cards,
  optional GitHub items, plan pins, and source coverage.
- `lco_operating_picture` powers cockpit-style views such as session maps,
  collaboration next steps, active-thread state, autonomy tick planning,
  GitHub operating items, and business pulse cards.

**Command layer for orchestrators**

- Exposes the same local registry through CLI commands, a Hermes-compatible
  MCP server, and an OpenClaw compatibility plugin.
- Gives normal agents a compact facade: prepared inbox, recent sessions, project
  digest, attention inbox, bounded expansion, describe, and Codex control dry
  run.
- Creates dry-run command packets for Codex start/resume/send/steer/interrupt
  actions so the target and action can be reviewed before live execution.
- For live Codex actions, the packet can include the exact target, action,
  message hash, and approval id your main agent should show before it moves.
- Pins the currently supported live Codex control path to `approvalPolicy=never`
  and a read-only, no-network sandbox; LCO does not inherit or widen a thread's
  ambient runtime permissions. Active-turn steer and interrupt fail closed
  unless the same-connection resume response proves that posture is already in
  effect.
- Adds hook-sidecar commands for closeout capture, state prep, compaction marker
  capture, and thread title aliases.

## Who It Is For

Use LCO if you:

- run Codex across many repos, customer projects, or product lanes
- use Hermes as your main local agent/operator
- want an MCP or OpenClaw orchestrator to manage Codex work
- need agents to hand off work without rereading massive transcripts
- want one place to ask "what is active, blocked, stale, finished, or ready?"
- want project digests and next-action briefs your agents can actually use

If you only run one short Codex session at a time, LCO may be more system than
you need. If Codex is becoming your day-to-day engineering workforce, this is
the memory layer that helps a main agent manage it.

## Install 🚀

Requirements:

- Node.js 22.5 or newer
- npm
- local Codex session files, usually under `~/.codex/sessions`
- Hermes if you want the primary supported agent path over stdio MCP
- OpenClaw Desktop/CLI if you want the compatibility plugin

Stable install:

```bash
npm install -g lossless-codex-orchestrator@latest
lco doctor
```

`lossless-codex-orchestrator` is the current published npm package name. The
deprecated compat package `lossless-openclaw-orchestrator` remains maintained
for existing automation and points at the same `lco` CLI and `lco-mcp-server`.
The historical `loo`, `loo-mcp-server`, and `LOO_*` env names remain maintained
compatibility aliases for at least two minor releases.

```bash
loo index codex "$HOME/.codex/sessions"
loo-mcp-server
```

Beta train, when you explicitly want the newest prerelease:

```bash
npm install -g lossless-codex-orchestrator@beta
```

Package channels:

- npm `latest` and GitHub Releases are the stable publication authorities.
- A version in the source tree is a candidate until those publication
  authorities show the same version; merged source alone is not release proof.
- `latest` is the stable public channel.
- `beta` is the active prerelease train.
- `next` is reserved for release candidates.
- npm dist-tags and GitHub Releases are the publication authorities; source
  version metadata, merged code, and local package smoke are not release proof.

If npm shows a version or dist-tag but install fails with a selector cutoff
error such as `ENOVERSIONS` or `ETARGET`, use the npm selector-drift tarball
fallback with raw npm commands a fresh shell can run:

```bash
tarball_url="$(npm view lossless-codex-orchestrator@latest dist.tarball)"
test -n "$tarball_url" && npm install -g "$tarball_url"
```

If the `ETARGET` message says the requested package version must have a publish
date before a specific time, check for a local npm `min-release-age` or `before`
pin before treating it as registry drift.

Full setup instructions live in [docs/SETUP.md](docs/SETUP.md).

## Set Up

Choose where LCO stores its local index. The default is already under
`~/.openclaw`, but setting it explicitly makes setup easier to inspect:

```bash
export LCO_DB_PATH="$HOME/.openclaw/lossless-openclaw-orchestrator/orchestrator.sqlite"
```

Index local Codex sessions:

```bash
lco index codex --max-files 500 "$HOME/.codex/sessions" "$HOME/.codex/archived_sessions"
```

The importer applies a 256 MB / 200,000-event per-file index cap so one
oversized JSONL cannot dominate a first run. If `lco index codex` reports
`codex_index_limited_files_skipped`, use `--max-bytes-per-file <bytes>` and
`--max-events-per-file <events>` only when you intentionally want to widen that
local indexing window.

The per-event content cache is local derived data used for deeper recall. If you
need to pause that cache or reclaim space, use:

```bash
export LCO_EVENT_CONTENT=disabled
lco maintenance --drop-event-content
```

Re-enable by unsetting `LCO_EVENT_CONTENT` and running `lco index codex` again.

Optional: allow recall from one or more OpenClaw LCM peer databases:

```bash
export LCO_LCM_DB_PATHS="$HOME/.openclaw/lcm.db"
```

Configured peers stay read-only. `lco index codex` and `lco find` materialize
public-safe advisory cards and inbox items from their summary DAGs without
copying peer rows into LCO source tables. Inspect peer readiness and integrity
with `lco doctor --peers`; peers are classified as ready, degraded, or
unavailable with omission reasons for missing tables, empty summaries, and
stale DAG links.

Check local readiness:

```bash
lco doctor --peers
lco onboard status --strict
```

## First Workflow 🧭

Find the work you remember. `lco find` runs a local incremental index pass on
first use, then searches titles, metadata, prepared cards, summaries, and
event-level content snippets:

```bash
lco find "billing bridge proposed plan"
```

Use JSON when you want the same result shape in scripts or agent harnesses:

```bash
lco find --json "billing bridge proposed plan"
```

For lower-level recall, use `lco search` for title/session-card discovery and
`lco grep` or `lco expand-query` for content-oriented recall.

Before preparing a drive plan, ask what changed since a public-safe cursor:

```bash
lco session-diff --cursor <cursor>
```

Build a bounded review-then-drive packet without running live control:

```bash
lco drive --dry-run --reviewer codex --driver claude --max-turns 3
```

When LCM peers are configured, inspect their read-only integrity posture with
`lco doctor --peers`. Claude targeting in 1.6 validates availability and
dry-run policy only; it does not provide Claude live control.

Describe a result:

```bash
lco describe codex_thread:<thread-id>
```

Expand a small brief:

```bash
lco expand-ref --profile brief --token-budget 1000 codex_thread:<thread-id>
```

Expand from a query when you do not know the ref yet:

```bash
lco expand-query --profile brief --token-budget 1000 "billing bridge"
```

For an agent or MCP client, start with the normal operator path:

| Step | Tool | What your agent gets |
| --- | --- | --- |
| 1 | `lco_find` | Fast public-safe session/content matches from the existing index; pass `index:true` only for an explicit synchronous refresh. |
| 2 | `lco_prepared_inbox` | The best starting view of work that needs attention. |
| 3 | `lco_describe_ref` | Details for a selected thread, card, leaf, or source ref. |
| 4 | `lco_expand_query` | A bounded brief when the exact ref is unknown. |
| 5 | `lco_recent_sessions` | Recent and active Codex work as compact cards. |
| 6 | `lco_attention_inbox` | Blocked, waiting, stale, approval-needed, or ready-for-review work. |
| 7 | `lco_project_digest` | A project-level handoff brief. |
| 8 | `lco_codex_control_route` | An expiring opaque reference for one daemon-owned Codex task, or an explicit Desktop-observation blocker. |
| 9 | `lco_codex_deliver` | A dry-run-first delivery that sends when idle or steers the matching active turn. |
| 10 | `lco_codex_control_dry_run` | A preview packet for a lower-level exact Codex action. |
| 11 | `lco_codex_resume_thread` | Resume a Codex thread after the dry-run packet is approved. |

The packaged agent playbook is
[skills/lossless-openclaw-orchestrator/SKILL.md](skills/lossless-openclaw-orchestrator/SKILL.md).

Naming note: `LCO` is the public product abbreviation. New user-facing examples
use `lco`, `lco-mcp-server`, and canonical `lco_*` tools. The historical `loo`,
`loo-mcp-server`, and `loo_*` names remain maintained compatibility aliases for
at least two minor releases.

## Works With 🔌

| Surface | Status | What to use today |
| --- | --- | --- |
| Codex local sessions | Stable | `lco index codex`, `lco search`, `lco describe`, and bounded expansion. |
| MCP clients | Stable | `lco-mcp-server` exposes the local tool registry over stdio. |
| Hermes | Primary supported agent path over stdio MCP; native adapter deferred | Mount `lco-mcp-server`, use the canonical `lco_*` tools, and run the Hermes smoke before release. |
| OpenClaw | Compatibility supported | Install the plugin and let your OpenClaw agent call the same `lco_*` tools. |
| Other MCP agents | Supported | Mount the standards-compliant stdio server; see [docs/HERMES_ADAPTER_BOUNDARY.md](docs/HERMES_ADAPTER_BOUNDARY.md). |

LCO is Hermes-first. Hermes uses the standards-compliant stdio MCP surface as
the primary supported agent path; OpenClaw remains a maintained compatibility
surface over the same runtime-neutral core. A native Hermes adapter is still a
separate, deferred claim.

Claude Code users who already run `codex-plugin-cc` can add LCO as a separate
recall companion:

```text
/plugin marketplace add 100yenadmin/Lossless-Codex-Orchestrator-LCO
/plugin install lco-recall@lco
```

The companion provides a user-invocable `find` skill for local recall and
leaves Codex command ownership with `codex-plugin-cc`.

## OpenClaw And MCP

Run the MCP server directly:

```bash
lco-mcp-server
```

Typical MCP client entry:

```json
{
  "mcpServers": {
    "lco": {
      "command": "lco-mcp-server"
    }
  }
}
```

Hermes configuration (`config.yaml`) is credential-free because LCO is a local
stdio server:

```yaml
mcp_servers:
  lco:
    command: lco-mcp-server
    enabled: true
    env:
      LCO_TOOL_PROFILE: standard
      LCO_CODEX_TRANSPORT: daemon
```

Omitting `LCO_DB_PATH` uses LCO's home-based default. If you set it explicitly,
use an expanded absolute path because Hermes does not shell-expand `~` inside an
environment-variable value.

`LCO_CODEX_TRANSPORT` defaults to `stdio` for compatibility. The opt-in
`daemon` mode connects only to the already-running local Codex managed daemon
through its Unix socket. LCO does not start or restart the daemon and does not
enable Codex Remote Control. Set `LCO_CODEX_DAEMON_SOCKET` only for an explicit
absolute local override; `LCO_CODEX_APP_SERVER_ARGS` remains stdio-only.

For remote operation, Eva should call `lco_codex_control_route` first. A
selected `app_server` route is safe for LCO daemon/CLI task control. A
`desktop_observation_required` result means Hermes must use its own
`computer_use` integration to identify and operate the Codex Desktop task;
LCO must not silently redirect that task to the managed daemon. Telegram
direction is user to Eva for instructions and Eva to the same user for results.
An operator should not type a bot message into Telegram as though it came from
Eva.

Before a release, verify the candidate without changing a Hermes profile:

```bash
lco qa-lab cli-mcp-smoke --evidence-dir /tmp/lco-package-smoke --package-version <package-version> --candidate-sha <candidate-sha> --cli-bin <candidate-prefix>/node_modules/.bin/lco --mcp-bin <candidate-prefix>/node_modules/.bin/lco-mcp-server --strict
lco hermes smoke --evidence-dir /tmp/lco-hermes-smoke --package-version <package-version> --candidate-sha <candidate-sha> --cli-bin <candidate-prefix>/node_modules/.bin/lco --mcp-bin <candidate-prefix>/node_modules/.bin/lco-mcp-server --strict
lco release hermes-readiness --evidence-dir /tmp/lco-hermes-readiness --package-version <package-version> --candidate-sha <candidate-sha> --hermes-smoke /tmp/lco-hermes-smoke/hermes-smoke.json --package-smoke /tmp/lco-package-smoke/cli-mcp-product-smoke.json --strict
```

The reports are public-safe candidate evidence. They do not prove publication,
an active Hermes profile install, or Eva runtime safety.

Install the OpenClaw plugin from npm:

```bash
openclaw plugins install lossless-codex-orchestrator@latest
openclaw plugins list --json
```

Smoke the OpenClaw path:

```bash
lco openclaw dogfood --profile lco-dogfood --install-source lossless-codex-orchestrator@latest --required-tool lco_doctor --required-tool lco_search_sessions --strict
lco openclaw tool-smoke --profile lco-dogfood --required-tool lco_doctor --required-tool lco_search_sessions --strict
```

Tool exposure can be narrowed with `LCO_TOOL_PROFILE=facade|standard|all`.
The default is `all`, preserving the full catalog. `facade` exposes the compact
operator path (`lco_*`) plus `loo_*` compatibility aliases; `standard` adds
workflow-detail tools.
Profile filtering affects tool listing and OpenClaw declarations only.

See [docs/OPENCLAW_PLUGIN.md](docs/OPENCLAW_PLUGIN.md) for the full OpenClaw
setup path.

## Privacy And Local Data

LCO reads local Codex session files and writes a local SQLite index. It is built
so agents can work from source refs, cards, summaries, and bounded briefs
instead of opening enormous raw transcript files by default.

For details, read [docs/PRIVACY.md](docs/PRIVACY.md) and
[docs/SAFE_SUMMARIES.md](docs/SAFE_SUMMARIES.md).

## Community And Contributing

LCO is meant to be easy to try and straightforward to improve. The public
contribution path is:

1. Read [CONTRIBUTING.md](CONTRIBUTING.md) for issue routing, validation,
   evidence, and agent-authored PR expectations.
2. Follow [docs/SETUP.md](docs/SETUP.md) for local install and first-run setup.
3. Use [AGENTS.md](AGENTS.md) when a coding agent is making or reviewing the
   change.
4. File a bug, docs bug, feature request, adapter request, protocol drift
   report, or command-control report through the GitHub issue forms.

Good first issue candidates are docs gaps, redacted fixture coverage, setup
diagnostics, issue-template improvements, and narrow CLI help fixes. Use the
adapter request form to request an adapter; describe the target app/runtime,
read/index path, command path, and sanitized evidence available to test it.

Never paste raw Codex transcripts, private SQLite databases, tokens, cookies,
connector URLs, or customer data into public issues or PRs. Use
[SECURITY.md](SECURITY.md) for private vulnerability reports and
[CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md) for community expectations.

## Architecture

- `packages/core`: local SQLite index, field-weighted search, source refs,
  summary leaves, prepared cards, inboxes, session descriptions, and bounded
  expansion
- `packages/mcp-server`: stdio MCP server exposing `lco_*` operator tools and
  `loo_*` compatibility aliases
- `packages/openclaw-plugin`: OpenClaw plugin entry and manifest surface
- `.codex-plugin/` and `hooks/`: Codex plugin manifest, Stop hook config, and
  wrapper for local thread title aliases
- `packages/cli`: `lco` CLI for setup, indexing, recall, smoke, eval, and
  release checks, with `loo` retained as a compatibility alias
- `packages/adapters`: Codex transport, audit, redaction, desktop-readiness, and
  adapter contracts
- `skills/`: packaged agent-facing OpenClaw playbook
- `evals/`: scenario and scorecard contracts
- `docs/`: setup, privacy, source authority, adapter docs, and maintainer guides

## Roadmap

Stable today: local Codex indexing, field-weighted search, prepared cards,
prepared inboxes, summary leaves, bounded expansion, project digests,
OpenClaw/MCP tools, and the dry-run command layer.

Since 1.2.x, the 1.2 prepared-state and summary-leaves lane has remained
shipped as part of the stable product line. That work added source-ref-backed
ranges, deterministic summary leaves, prepared cards, prepared inbox items,
watcher observations, attention queue items, and hook capture so an OpenClaw or
main orchestration agent can start from compact prepared state instead of
rereading huge Codex transcripts. The historical architecture handoff remains in
[docs/sprints/brief-lco-1.2-prepared-state-summary-leaves-2026-07-03.md](docs/sprints/brief-lco-1.2-prepared-state-summary-leaves-2026-07-03.md).

The 1.4 identity work makes `lco`, `lco-mcp-server`, `lco_*`, and `LCO_*` the
canonical command and tool names, with `loo`, `loo-mcp-server`, `loo_*`, and
`LOO_*` retained as compatibility aliases for at least two minor releases. The
current npm package is `lossless-codex-orchestrator`; the deprecated compat
package `lossless-openclaw-orchestrator` remains maintained for existing
automation.

For the full product direction, read [VISION.md](VISION.md).

## Development

```bash
npm install
npm run build
npm test
npm run check
```

The test suite uses redacted fixtures and Node's built-in test runner. Heavy
validation is expected to run in CI; local development should use focused tests
first.

For contributor workflow details, use [CONTRIBUTING.md](CONTRIBUTING.md).

## License

Current source and current npm-published versions are source-available under the
[PolyForm Noncommercial License 1.0.0](LICENSE). Noncommercial use is permitted
under those terms.

**Using LCO yourself — including at work — is fine and encouraged.** Embedding LCO inside a product or
service that you sell requires a commercial license (standard terms: 5% of the embedding product's
pricing, or negotiated). Plain-language guide:
[docs/COMMERCIAL_LICENSING.md](docs/COMMERCIAL_LICENSING.md) — or just email
**support@electricsheephq.com**.