outfitter-dispatch 0.8.1

Local control plane for orchestrating Codex agent lanes over the Codex App Server.

dispatch

Local control plane for orchestrating Codex agent lanes over the Codex App Server. One authored contract per operation, projected to CLI + MCP (+ remote later) with no drift.

Quick Start

Install the CLI from PyPI:

uv tool install outfitter-dispatch
dispatch --help
dispatchd --help
dispatch doctor
dispatch up --json
dispatch down --json

From a source checkout:

uv sync
uv run dispatch --help
uv run dispatch doctor --no-app-server
uv run dispatch models --no-refresh
uv run dispatch up --json
uv run dispatch daemon status

Create an owned managed thread, send it work, and inspect the daemon:

uv run dispatch new \
  --name docs \
  --cwd /path/to/dispatch \
  --goal "Finish the docs review." \
  --text "Please summarize the current stack state."
uv run dispatch list
uv run dispatch get <dispatch-ref>
uv run dispatch tail <dispatch-ref> --limit 20
uv run dispatch daemon log --limit 10
uv run dispatch down --json

For durable or parallel launches, point new at a launch packet directory and preview it without side effects: dispatch new --name lane-a --cwd /repo --packet ./packet --dry-run --json, then --stage all to write durable session files under .agents/sessions/<ref>/. See docs/usage/README.md for packet layout, file/stdin inputs, and staging.

Use owned managed threads for turn-writing work. Existing desktop Codex threads can be attached as managed threads, but ADR-0005 blocks turn-writing and history-mutating commands such as send, stop, goal set, and goal clear on attached lanes by default. A local operator can explicitly opt in with [policy] allow_attached_writes = true in ~/.dispatch/config.toml; list --json and get --json expose writable, capabilities, and write_locked_reason so scripts can tell which lanes can receive writes. Every managed thread has a dispatch-local ref; full Codex thread UUIDs remain accepted everywhere. Titles and @handles are mutable convenience labels, not stable identity. Metadata lifecycle actions (rename, archive, restore) can target managed refs or raw unmanaged Codex thread ids, and search can span both. Attach is metadata-only by default; use dispatch sync <selector> when you want dispatch to refresh its local indexed view of an attached thread. If <selector> is a raw unmanaged Codex thread id, sync first registers it as an attached read/metadata-managed lane, then refreshes the index. dispatch list --unmanaged --archived shows archived Codex sessions before you decide whether to sync or restore them. Bare dispatch history reads Dispatch's local index only; selector-scoped transcript reads through tail, history, or transcript-inclusive get use App Server thread/read(includeTurns:true) as the canonical source and backfill Dispatch's normalized local history index for that one thread.

History capture is configurable in ~/.dispatch/config.toml. The default standard mode captures operational facts and bounded searchable history metadata while keeping raw provider payloads gated. Live App Server events are stored as compact summaries; transcript reads index bounded turns, item text, tool names, and file/thread refs without retaining raw item payloads unless the retention policy allows it. Minimal capture keeps turn-level state but skips item-level transcript rows. Bare history overview is a local indexed summary. Normal selector-scoped history item/tool/file views render from the normalized index after refreshing one thread from the App Server, while history --raw remains a live raw-payload inspection path. Use debug only for development with bounded temp state; debug retention can store bounded raw provider event and item payloads with truncation markers for reducer/search diagnosis:

[history]
capture = "standard" # minimal | standard | debug
raw_payload_retention = "debug" # off | errors | debug | all
max_text_bytes = 8192
max_payload_bytes = 65536

dispatch doctor reports the active capture mode and warns when debug/raw retention is enabled.

new reports whether the first message was accepted by the App Server, not whether assistant work completed. Use get to inspect the latest turn state and persisted App Server errors, or watch for a bounded live event sample. Slash commands in --text are plain text; use --goal when creating a native App Server goal. Use dispatch models before pinning model or service-tier presets; Dispatch resolves aliases such as fast from the live App Server model catalog and keeps omitted model/tier values on Codex defaults.

For the operator guide, CLI/MCP examples, triggers, and plugin setup, start at docs/usage/README.md.

Start troubleshooting with dispatch doctor. It checks PATH visibility, the Codex CLI and auth footprint, daemon socket/pidfile state, registry schema/integrity, packaged skills/plugin assets, and a low-risk Codex App Server initialize smoke. If doctor reports an old registry schema, stop the daemon and run dispatch registry migrate before starting it again.

Agent And Plugin Support

This repo ships first-party skills in skills/:

• skills/dispatch/SKILL.md teaches agents how to operate dispatch safely.
• skills/dm/SKILL.md is the dispatch-backed "dispatch message" workflow for short inter-lane messages.

The workspace-local Codex plugin bundle lives at plugins/dispatch/, with a marketplace entry in .agents/plugins/marketplace.json. Restart Codex if the plugin does not appear immediately.

Project Docs

• docs/development/design.md - architecture and design notes.
• docs/adrs/ - accepted architecture decisions.
• docs/research/ - verified Codex App Server findings.
• .agents/plans/v0/RETRO.md - v0 execution ledger and verification record.

For contributors, AGENTS.md is the canonical fieldguide.