Agent 0 Roster

Agent 0 roster

Agent 0 coordinates the selected project's plan, canonical tasks, delegated work, integration, and acceptance. The read-only roster is one navigation view of registered agents, AUTOBOT references, BOT/DOT instances, and recorded CLI channels. It does not start an agent, choose a provider, run a task, or grant access. WORK/TASKS.json, role registries, Agent Beta records, and the Access Matrix JSON remain authoritative for their own facts.

Open the roster

From the directory containing .SYSTEMX:

bash .SYSTEMX/SYSTEMX.sh root roster
bash .SYSTEMX/SYSTEMX.sh root roster --help

Windows PowerShell uses pwsh -File .SYSTEMX/SYSTEMX.ps1 root roster. A managed installation can use systemx run --target "/path/to/project" --offline -- root roster. To inspect a registered child, use the projects route shown by projects root roster --help; a direct root command addresses the outer scope.

The result reports a sourceRevision, counts, rows, access-decision references, cliProviderOptions, and setupCommands. The blank template has 20 stock actor/pattern rows, Agent 0 registered, and zero BOT, CLI-channel, or access decision records. A row shows actor ID, kind, registration state, source record, workspace, provider alias, task count, decision IDs, and reviewed-intent, denial, and gap counts. BOT rows expose the provider name only, without a provider identity or secret. Native subagents appear as a pattern unless a selected-scope decision names a specific instance, in which case the row is marked decision-only. Explicit human or service access actors also appear only as decision-only rows; the roster does not invent their identities. externalRightsVerified is false for every row. These are review pointers, not a second task ledger or live provider probe. Re-read after registration, a task change, or an access decision; a saved response can become stale.

Use --actor ACTOR for one exact actor or recorded CLI-channel ID. --offset and --limit page roster rows; --decision-offset and --decision-limit page scoped access decisions separately. Limits are 1–50 rows per page. The child route is:

bash .SYSTEMX/SYSTEMX.sh projects root roster --project NAME

Open the menu and workbook tab

The local interactive menu exposes option 33, “Agent 0 agent, AUTOBOT, BOT, and CLI channel roster.” It calls the same read-only outer-scope root roster command. The menu does not initialize or activate a role.

The checked-in stock Access Matrix workbook has five tabs: Access decisions (with 20 stock role rows), Agenda options, Settings catalog, Decision rules, and Agent 0 roster. A selected-scope agenda workbook-export generates six formula-free sheets, separating Role defaults from Access decisions and adding the same Agent 0 roster tab. The generator derives it from canonical JSON. Its 13 columns are actorId, kind, displayName, registrationState, sourceRecord, workspace, providerAlias, taskCount, accessDecisionIds, reviewedIntentCount, deniedCount, gapCount, and externalRightsVerified. Preview bash .SYSTEMX/SYSTEMX.sh agenda workbook-export, then apply bash .SYSTEMX/SYSTEMX.sh agenda workbook-export --expected-revision REVISION_FROM_PREVIEW --apply with the returned sourceRevision to write the formula-free workbook under ignored local/. agenda workbook-check reports whether it still matches the selected source; a registration or decision change makes the previous tab stale. The checked-in workbook is a stock catalog view, not a live project roster.

The checked-in five-sheet stock workbook and six-sheet generated export were checked for row parity and formula absence; a roster registration or decision change changes the generated workbook revision and makes the older export stale.

Choose a role lane

Lane	Existing setup or inspection path	Boundary
Agent 0	root roster, root packet, canonical task commands	Coordinates and accepts; a roster row does not dispatch work.
Agent X, Y, Z	roles-init preview/apply, then each role's commands	Registration adds records, not running processes.
Agent Alpha	autobot init preview/apply; agenda catalog for fixed settings	Routes selected references; Agenda choices start unset.
Agent Beta	agent-beta setup and bot-add preview/apply	A named BOT/DOT workspace starts offline.
Agent1	agent-1 setup preview/apply and advisory check	Does not certify a device or install a daemon.
Native subagent	The chosen IDE or harness	The host assigns runtime identity and permissions.

A reviewed-intent Access Matrix decision remains recorded intent; identityVerified, accessGranted, and runtimeStarted stay false without independent external evidence. See Agent 0 and subagents, Access Matrix, and task workflow.

Configure external CLI providers separately

The installed systemx command manages template lifecycle. The project-local .SYSTEMX/SYSTEMX.sh or .SYSTEMX/SYSTEMX.ps1 launches selected record commands. Codex, Copilot, Muse Code, and any configured Agent Y CLI are separate external programs with their own installation, account, permissions, model availability, and cost. GitHub CLI gh is repository/PR tooling, not an AI provider. Several CLIs may be installed at once; the roster selects and authenticates none of them.

bash .SYSTEMX/SYSTEMX.sh agent-y providers
bash .SYSTEMX/SYSTEMX.sh onboard plan
bash .SYSTEMX/SYSTEMX.sh agent-y prepare --help

agent-y providers detects installed names, not usable login or permission. onboard plan shows five onboarding choices (account context, surface, provider, credential consumer, and credential class), the Agent1 open-question count, and the credential-consumer boundary without running a provider. Grok Build CLI, Grok Bot, OpenAI DOT, personal Muse, Amazon Kiro, AWS CLI, Codex SDK, and the six documented manual coding CLIs need separate host checks. The broader onboarding catalog is navigation, not a set of registered roster workers or Agent Y execution adapters. The roster does not claim that any online account is connected. See provider and harness onboarding.

For an IDE or harness that needs a bounded local readback, see Agent 0 and toolchain direct I/O. It covers project CLI, the fixed JSON-lines stdio methods, and host integration targets without claiming an installed MCP/ACP adapter or native IDE plugin. For each bounded task, review the selected executable, account, project scope, repository, write paths, timeout, branch/PR policy, and expected evidence before preparing it. Agent Y pins one provider for that task and checks it again before execution. Native app subagents and Agent Beta online references have different boundaries; the roster does not turn either into an Agent Y run. See CLI and Tools and Agent Y.