Skip to content

Terms and command entrypoints

This appendix routes readers; it does not replace the LoopX CLI reference. Run loopx <command> --help for complete arguments in the installed version.

Core terms

TermMeaning in this book
AgentExecutor that plans and performs one bounded action in a Host/runtime
HostProduct or runtime that owns sessions, model turns, and wake-up surfaces
GoalLong-running project outcome and state boundary identified by a stable goal_id
Agent identityPeer or lane identified by agent_id; it is not the Goal and does not prove the Host
VisionA bounded execution-routing contract for one agent_id, including role scope, direction, acceptance summary, and replan trigger
AcceptanceObservable conditions that prove the Goal is complete
TodoSchedulable work item with identity
FrontierTodos currently runnable after dependency, Gate, capability, and boundary checks
ClaimSoft ownership of a Todo
LeaseTime-bound exclusive reservation that prevents conflicting execution
GateBlocking decision with explicit scope and authority
EvidenceVerifiable material that supports a judgment
ReceiptDurable record of an accepted action or lifecycle transition
ProjectionRead model derived from canonical state
QuotaContract that decides whether a turn may run and records validated spend
MonitorTodo that observes an external condition on a cadence and advances only on material change
CapabilityCaller-facing outcome contract
ProviderImplementation or external-system caller that returns a bounded result
ExtensionInstallation, activation, upgrade, rollback, and compatibility lifecycle for a Provider or package
KernelCore that accepts transitions and owns durable control-plane state

Core protocol index

Choose a protocol from the developer job rather than reading the full directory alphabetically:

ProblemStart with
/loopx <goal text>, Goal selection, fresh Agent identity, and Host activationloopx_goal_command_v0
Long-running Agent sources, projections, concurrent lanes, and lifecyclelong_horizon_agent_state_protocol_v0
Agent-scoped chronology before replan or handoffagent_scoped_evidence_ledger_v0
Canonical events, replay, and privacyevent_sourced_state_contract_v0
Typed read model over the active-state workbenchactive_state_structured_projection_v0
Todo, Gate, dependency, and handoff graphtask_graph_projection_v0
Gate coverage and scoped authoritydecision_scope_v0
Per-Agent Vision and replangoal_vision_replan_contract_v0
Equal peers, claims, and continuationpeer_agent_runtime_v1
One governed execution transaction, currently experimentalloopx_turn_v0
Opt-in bounded projection for an already-arbitrated decisionturn_envelope_v0
Host capability, controlled write, and fallbackhost_integration_surface_v0
Read-only first-screen projection for a session runtimesession_runtime_loopx_projection_v0
Controlled session-runtime metadata writeback, currently draftsession_runtime_controlled_writeback_v0
Revision, idempotency, and local write correctnesslocal_state_write_correctness_v0

Use the official Protocol Contracts index for the complete current set.

Common read-only entrypoints

bash
loopx doctor
loopx registry
loopx status
loopx todo list --goal-id <goal-id>
loopx history --goal-id <goal-id>
loopx evidence-log --goal-id <goal-id> --agent-id <agent-id> --thin --limit 30
loopx quota should-run --goal-id <goal-id> --agent-id <agent-id>
loopx extension list --format json

Project onboarding entrypoints

bash
loopx connect

loopx start-goal --guided --project . \
  --goal-id <goal-id> \
  --agent-id <agent-id> \
  --goal-text "<goal text>" \
  --host-surface codex-app

loopx start-goal --guided --project . \
  --goal-id <goal-id> \
  --agent-id <agent-id> \
  --goal-text "<goal text>" \
  --host-surface codex-cli-tui

On first run, omit --goal-id, --agent-id, or --host-surface to receive the corresponding read-only Goal, fresh-Agent, or Host selection Gate. Rerun the exact command provided by the packet. Do not infer a Goal from similar text or take over the only existing Agent automatically. Register a new identity with a register-agent --goal-id <goal-id> --agent-id <new-agent-id> preview followed by atomic --execute; use an existing identity only for an explicitly authorized takeover.

Safe upgrade runbook

For a no-clone installation, loopx update is the primary path. Do not overwrite a release snapshot by hand:

Run loopx update --check for a read-only freshness check, then loopx update --dry-run for the install preview. Neither command installs.

A normal upgrade needs only:

bash
loopx update --check
loopx update --dry-run
loopx update --execute
loopx doctor

Use the full flow below when you need pre-upgrade evidence, Host or Extension migration checks, or a rollback path.

bash
# 1. Record current facts
command -v loopx
loopx --version
loopx --format json doctor > /tmp/loopx-doctor-before.json

# 2. Inspect stable ref, freshness, and the recommendation
loopx update --check

# 3. Preview the ref, release id, and rollback target
loopx update --dry-run

# 4. Run the archive installer and post-update doctor
loopx update --execute

# 5. Recheck commands, skills, Host integration, and project state
loopx --version
loopx doctor
loopx slash-commands
loopx slash-commands --install
loopx status

The public stable ref is the default source. --ref main is a maintainer or development qualification path, not the ordinary user default. update --execute installs a release snapshot and runs doctor; a successful exit does not prove that every Host automation, Goal migration, or Extension Provider is updated.

Validate the surfaces you use:

  • loopx doctor: wrapper, release manifest, Python import, skill delivery, and Host integration;
  • loopx slash-commands --install: updates only LoopX-managed command files and skips user-owned collisions;
  • loopx quota should-run or loopx upgrade-plan: peer-runtime and heartbeat-prompt migrations;
  • loopx extension list plus executed extension doctor: readiness for each active revision;
  • loopx status and history: registry, Goal, Todo, and projection continuity.

Before a risky migration, scheduler change, or runtime repair, preview and create a private local backup:

bash
loopx backup-state --project .
loopx backup-state --project . --execute

The archive contains local runtime and project state. It is private recovery material and must not be committed or published.

When a new release blocks normal work, inspect current loopx update --help, then select a recorded release id or use:

bash
loopx update --rollback previous
loopx doctor

Rollback restores the LoopX release snapshot only. Project state already written by the new version, external effects, and separately installed Extension packages may need their own migrations or rollback. Do not describe wrapper rollback as whole-system rollback.

Scheduler convergence entrypoint

When a Codex App packet reports stateful_backoff.apply_needed=true, have the Host apply recommended_rrule, read back the actual result, and then run the packet's full ack_hint.cli_args. The current route is typically:

bash
loopx quota scheduler-ack-current <packet-bound-args...>

After an apply failure or timeout, do not ACK; run failure_hint.cli_args once. When apply_needed=false and ack_needed=true, exact Host readback already matches the target cadence, so skip the no-op update and run the bound ACK. Proposal, Host apply, readback, and ACK are all required for convergence, and cadence changes do not consume delivery spend.

Extension lifecycle entrypoints

bash
loopx extension init <extension-id>
loopx extension install --manifest <extension.toml>
loopx extension doctor <extension-id>
loopx extension run <extension-id> --input-json <request.json>
loopx extension disable <extension-id>
loopx extension enable <extension-id>
loopx extension upgrade --manifest <extension.toml>
loopx extension rollback <extension-id>

Lifecycle commands normally preview by default. Inspect current --help and add --execute only when you intend to mutate state or invoke the Provider.

Source contribution entrypoints

Official sources