docs(fleet): add operator configuration guide

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

docs/fleet/concepts/desired-vs-observed-state.md

@@ -0,0 +1,42 @@
+# Desired, Derived, and Observed Fleet State
+
+## One writable authority
+
+The canonical local v2 roster at `<MOSAIC_HOME>/fleet/roster.yaml` is desired state. Membership, stable identity, class, runtime/provider/model selection, launch policy, enablement, and persisted `running`/`stopped` intent are written only through generation-guarded roster mutations.
+
+Derived projections are reproducible consequences of that authority:
+
+- `<name>.env.generated`;
+- exact roster-named tmux sessions on the configured socket after reconciliation;
+- systemd service targets managed by installation/reconciliation.
+
+Current systemd unit enablement is not yet lifecycle-conformant at boot: installation can enable every
+agent unit, and the launcher projection does not carry `enabled` or `desired_state`. Therefore reboot
+preservation for stopped/disabled agents remains an FCM-M3-002 acceptance hold, not a guaranteed
+projection behavior.
+
+Observed evidence available to current roster-v2 status commands includes systemd active state, tmux
+presence, holder ownership, and unmanaged sessions. Heartbeat files are observational in the wider fleet,
+but roster-v2 `status`, `doctor`, and `verify` do not currently read them. Observation never writes back
+to the roster.
+
+## Generation and ownership
+
+`generation` is a positive integer concurrency fence. A mutating request must provide the current value. Successful changed CRUD increments it exactly once; stale or concurrent writers fail before mutation. Apply/reconcile rereads the canonical roster under a private exclusive lock and uses only that generation and content for effects.
+
+Ownership is exact, never fuzzy. Reconciliation is limited to roster names, the configured socket, the exact holder session, a private installation identity, and private managed paths. An ownership mismatch, unmanaged session, unsafe path, stale generation, or ambiguous lock fails closed.
+
+## Drift decisions
+
+| Observation                              | Interpretation           | Safe response                                                          |
+| ---------------------------------------- | ------------------------ | ---------------------------------------------------------------------- |
+| Generated file differs or is missing     | Derived projection drift | Review `apply --dry-run`; regenerate from the roster.                  |
+| Desired `running`, exact session missing | `missing-session`        | Diagnose ownership/runtime, then reconcile if safe.                    |
+| Desired `stopped`, exact session present | `unexpected-session`     | Inspect; reconciliation may stop only the proven roster target.        |
+| Disabled agent running                   | `disabled-running`       | Inspect; disabled state wins during explicit safe reconciliation.      |
+| Unknown session on the configured socket | Unmanaged state          | Report only. Do not adopt, rename, or kill it.                         |
+| Heartbeat stale in the wider fleet       | Liveness evidence        | Diagnose separately; current roster-v2 status does not read heartbeat. |
+
+`status` and `doctor` classify. `verify` is also observational but exits non-zero when ownership, drift, or unmanaged-state checks fail. `plan`/`apply --dry-run` validates proposed projection and lifecycle work without mutation. `apply` and `reconcile` converge only after all preconditions pass.
+
+A partial projection failure does not roll the roster back. Treat the committed roster as authority and regenerate. A lifecycle failure after projection completion preserves both roster and projections for inspection. Sensitive legacy values are never printed; diagnostics are bounded to stable codes, key names where applicable, and hashes.

docs/fleet/concepts/generated-env-launch-chain.md

@@ -0,0 +1,23 @@
+# Generated Environment Launch Chain
+
+The launcher consumes validated data, not shell configuration.
+
+1. Read and validate the canonical roster.
+2. Render deterministic `<name>.env.generated` data from that roster.
+3. Parse optional `<name>.env.local` through a strict allowlist.
+4. Reject generated-key shadowing, unknown or sensitive-looking keys, unsafe paths/values, duplicates, malformed lines, shell syntax, and command overrides.
+5. Derive the runtime command from validated runtime/model/reasoning data.
+6. Target only the exact configured tmux socket and roster session after ownership checks.
+
+## File precedence and ownership
+
+| File              | Owner                    | Use                                                                         |
+| ----------------- | ------------------------ | --------------------------------------------------------------------------- |
+| `.env.generated`  | Mosaic projection writer | Complete deterministic roster projection. Rebuild; do not edit.             |
+| `.env.local`      | Operator                 | Optional, private, strict machine-local data. Cannot shadow generated keys. |
+| `.env`            | Legacy input             | One-time migration input only; never launch authority.                      |
+| `.env.quarantine` | Private quarantine       | Retained unsafe legacy evidence; never loaded by the launcher.              |
+
+Neither systemd nor the launcher sources these files. No `eval`, shell expansion, arbitrary `MOSAIC_AGENT_COMMAND`, channel, or secret-reference compatibility path exists. Safe legacy generated keys are regenerated, allowed local keys are relocated, and unsafe material is quarantined.
+
+Diagnostics never expose the rejected value, credential material, or command text. They are bounded to stable rule code, key name where safe, and SHA-256 content identity. See [generated environment reference](../reference/generated-env-boundary.md) and [quarantine operations](../operations/env-quarantine.md).

docs/fleet/concepts/identity-class-runtime.md

@@ -0,0 +1,20 @@
+# Fleet Identity, Class, and Runtime
+
+Each roster field has one job. Do not use names or model strings as authority shortcuts.
+
+| Concern                 | Field                           | Contract                                                                                     |
+| ----------------------- | ------------------------------- | -------------------------------------------------------------------------------------------- |
+| Stable machine identity | `agents[].name`                 | Unique, immutable mutation target and exact service/session name.                            |
+| Display identity        | `agents[].alias`                | Human-facing label only; may be changed and grants no authority.                             |
+| Behavioral contract     | `agents[].class`                | Resolves through the shared baseline plus `roles.local` persona library.                     |
+| Tool boundary           | `agents[].tool_policy`          | Must match protected canonical classes; cannot independently grant authority.                |
+| Harness                 | `agents[].runtime`              | One of `claude`, `codex`, `opencode`, or `pi`, declared in `runtimes`.                       |
+| Backend selection       | `agents[].provider` and `model` | Explicit non-empty data; capability validity is not inferred from the display name or class. |
+| Effort                  | `agents[].reasoning`            | `low`, `medium`, or `high`.                                                                  |
+| Local placement         | `working_directory`             | Explicit safe local work path; not remote placement authority.                               |
+
+Tess and Ultron are conventional instance/display names only. They are not products, required machine identities, role aliases, or authority-bearing classes. A configurable interaction instance uses `class: interaction`; a configurable validation instance uses `class: validator`. Any stable name and alias satisfying the structural contract may be used.
+
+Class aliases are deliberately narrow: `implementer → code`, `reviewer → review`, and `operator-interaction → interaction`. No runtime, provider, model, persona prose, or instance name changes this mapping. See [role classes](../reference/role-classes.md) and the [validated generic example](../examples/roster-v2.yaml).
+
+Roster v2 is local-only. It contains no host/SSH placement, connector, channel, secret-reference, arbitrary-command, per-agent socket, or gateway mapping fields. Those concerns require separate requirements and threat models.

docs/fleet/concepts/role-authority-and-leases.md

@@ -0,0 +1,22 @@
+# Fleet Role Authority and Leases
+
+Role content describes behavior; protected authority is immutable code metadata derived only from the canonical class.
+
+## Required workstream classes
+
+`code`, `review`, `validator`, `orchestrator`, `team-leader`, `enhancer`, and `interaction` are required FCM classes. `merge-gate` is additionally protected because it remains the sole approve-to-land and merge authority.
+
+| Class                                        | Authority                                         | Boundary                                                                               |
+| -------------------------------------------- | ------------------------------------------------- | -------------------------------------------------------------------------------------- |
+| `merge-gate`                                 | Approve-to-land and merge                         | Sole merge authority.                                                                  |
+| `validator`                                  | Issue independent validation evidence/certificate | Never approves landing or merges.                                                      |
+| `orchestrator`                               | Orchestrate topology and issue bounded leases     | Does not gain merge authority.                                                         |
+| `team-leader`                                | Use explicitly leased capacity                    | Cannot issue leases or mutate roster, credentials, topology authority, or merge state. |
+| `interaction`                                | Receive requests and report status                | Cannot orchestrate, issue leases, mutate configuration, or merge.                      |
+| `code`, `review`, `enhancer`, custom classes | No protected authority by default                 | Persona prose cannot grant protected powers.                                           |
+
+A lease is capacity authorization from an orchestrator, not ownership. It must identify a bounded task or period and does not alter the leased agent's roster identity, role contract, credentials, authority, or persisted lifecycle. Expiry/revocation returns capacity; it does not rewrite the roster.
+
+Semantic validation rejects protected class/tool-policy mismatch in either direction. An instance named Ultron with `class: validator` remains validation-only. An instance named Tess with `class: interaction` remains request/status-only. Renaming either instance changes no authority.
+
+For resolver layering and safe customization, see [role classes](../reference/role-classes.md) and [customize roles](../how-to/customize-roles.md).