[issue #98] Unify Saihai naming and align setup/classify docs with implementation

README.ja.md

@@ -1,8 +1,8 @@
-# Sahai
+# Saihai
 
 [English](README.md) | [日本語](README.ja.md)
 
-Sahai は、プロンプトを実行権限として扱うのではなく、typed artifact、明示的な承認、durable state、監査可能な evidence を通じて AI エージェントの作業を実行するためのローカル orchestrator / organization runtime リポジトリです。
+Saihai は、プロンプトを実行権限として扱うのではなく、typed artifact、明示的な承認、durable state、監査可能な evidence を通じて AI エージェントの作業を実行するためのローカル orchestrator / organization runtime リポジトリです。
 
 このリポジトリには Agent-Teams-Viewer（ATV）由来のローカル status viewer も残っていますが、現在の主要な product surface は deterministic frontdoor、durable workflow run、制約された main-agent bridge、typed report / evidence gate です。pre-release の記録では、historical artifact や compatibility alias を指す際に ATV の名称が残っている場合があります。
 
@@ -43,7 +43,8 @@ daemon scheduling、tmux worker execution、package distribution、automatic pub
 
 ```sh
 python3 scripts/setup_directory_paths.py --help
-# Supply all nine required directory options, then validate the catalog.
+# configuration guide に示す8個の directory option を指定してから catalog を
+# 検証する。SAIHAI_ROOT の default は primary checkout。
 python3 scripts/setup_directory_paths.py --check
 ```
 
@@ -64,7 +65,7 @@ setup command は非破壊的で、owner-only file を書き込みます。proce
 
 | 機能 | 現在の挙動 |
 |---|---|
-| Prompt classification | `scripts/configure_organization.py classify` と `/api/decide` が作業を `fast`、`strict`、`maintenance` に分類する。どの mode でも適用される task record と Vault record が必要 |
+| Prompt classification | `scripts/configure_organization.py classify` と `/api/decide` は、`fast` / `strict` の execution mode と、`enabled` / `disabled` / `maintenance` の organization state を別々に返す。どの mode / state でも適用される task record と Vault record が必要 |
 | Workflow selection | `workflow_selector.py` が typed classification を active workflow template へ deterministic に対応付ける。raw prompt は selection authority にならない |
 | Frontdoor proposal | prompt 起点の request は `proposed` または `waiting_human` まで進むか、`blocked` として fail closed する。`propose` は approved activation を生成せず、workflow run も作成しない |
 | Approval | `approve` は proposal digest から導出した challenge を検証する。許可される activation source は `human_ui`、`manual_cli`、`orchestrator-start`、trusted execution principal は `human_operator`、`manual_operator`、`orchestrator_start`。narrow CLI の default は `human_operator` / `human-ui` / `local_ui` |
@@ -204,8 +205,8 @@ provider CLI の1回の実行は既定30分で、1秒から24時間まで設定
 ```sh
 python3 scripts/configure_organization.py status
 python3 scripts/configure_organization.py runtime-paths
-python3 scripts/configure_organization.py classify --prompt "Review the latest forecast"
-AGENT_ORG_MAINTENANCE=1 python3 scripts/configure_organization.py classify --prompt "Repair a hook"
+python3 scripts/configure_organization.py classify --prompt "Review the latest forecast" --mode strict --organization-state enabled
+python3 scripts/configure_organization.py classify --prompt "Repair a hook" --organization-state maintenance
 python3 scripts/configure_organization.py validate-all
 python3 scripts/configure_organization.py workflow-selector validate-contracts
 python3 scripts/configure_organization.py workflow-frontdoor --help
@@ -215,7 +216,7 @@ python3 scripts/configure_organization.py workflow-frontdoor --help
 |---|---|
 | `status` | organization settings、role/policy count、repository root を JSON で出力する |
 | `runtime-paths` | ITB runtime、workflow selector/frontdoor/server、operator CLI、registry mirror を検証する |
-| `classify` | prompt を `fast`、`strict`、`maintenance` に分類する |
+| `classify` | optional な `--mode {fast,strict}` と、独立した `--organization-state {enabled,disabled,maintenance}` control を使って prompt を分類する |
 | `validate-all` | offline suite、contract validation、Python compile check を実行する |
 | `workflow-selector` | workflow contract を検証し、deterministic selection と activation-envelope operation を行う |
 | `workflow-frontdoor` | host-owned な frontdoor / recovery surface 全体を提供する |
@@ -269,15 +270,15 @@ Codex profile、必須のadministrator deployment、canary、worker policy-domai
 
 ### Scoped worker backend
 
-live scoped-worker backend は、host operator が以下の asset を手動設定するまで fail closed します。Sahai は key や credential を生成しません。
+live scoped-worker backend は、host operator が以下の asset を手動設定するまで fail closed します。Saihai は key や credential を生成しません。
 
 executor implementationはshippedですが、live `managed_worker` claimはsuppressされています。Codex 0.144.1ではsame-rootfsなlocal `git push`の絶対拒否を証明できません。currentなworker commissioningはevidenceを記録できても、`commission-seal`が`worker_denial_facts_not_promotable`でfail closedします。currentな`external_mutation`、`git_commit`、`git_push`、`credential_access` factは`result=fail`でhost observationもinconclusiveです。特に`workspace_profile_and_network_disabled_not_same_rootfs_isolation`と`dedicated_auth_deny_configured_not_mechanically_proven`は機械的なdeny evidenceではなく、明示的なnon-claimです。active worker claimには別のisolated policy domainで得たevidenceが必要です。shipped local action gatewayにはそのdomainへのautomatic transportがないため、current releaseは自動接続済みのend-to-end worker systemではありません。
 
 | Environment variable | 用途 |
 |---|---|
 | `SAIHAI_SCOPED_EXECUTOR_KEY_FILE` | regular、non-symlink、`0600` file に保存された32 bytes以上の capability HMAC key |
 | `SAIHAI_SCOPED_WORKTREE_ROOT` | host が task/run-bound worktree path を導出する canonical root |
-| `SAIHAI_SCOPED_REPO_ROOT` | host-owned absolute repository path。default は Sahai repository root |
+| `SAIHAI_SCOPED_REPO_ROOT` | host-owned absolute repository path。default は Saihai repository root |
 | `SAIHAI_SCOPED_CODEX_EXECUTABLE` | digest が work order と capability に bind される absolute pinned Codex CLI path。group/world-writable binary は拒否される |
 | `SAIHAI_SCOPED_CODEX_HOME` | dedicated worker runtime/auth root。main-agent profile は継承しない |
 | `SAIHAI_ENABLE_SCOPED_WORKER_LIVE=1` | 明示的な live-execution gate。未設定時は deterministic fake harness だけが利用可能 |
@@ -369,7 +370,7 @@ ITB session discovery は `~/.claude/state/itb` と `~/.codex/state/itb` を読
 | `GET /api/org?session=<id>` | team role state、active task、busy count |
 | `GET /api/role?session=<id>&role=<role_id>` | role metadata、inbox、latest report、provider evidence |
 | `GET /api/config` | organization settings と role/policy index |
-| `GET /api/decide?prompt=<text>` | `fast`、`strict`、`maintenance` classification |
+| `GET /api/decide?prompt=<text>` | `fast` / `strict` mode と、独立した `enabled` / `disabled` / `maintenance` organization state を返す decision |
 | `GET /api/workflow-runs?session=<id>&task=<id>&state=<state>` | thin read-only workflow-run summary |
 | `GET /api/workflow-run?session=<id>&run=<id>` | work order、report、provider evidence、transition metadata |
 | `GET /api/workflow-lock` | 設定済み orchestrator root ごとの global workflow-lock status |

README.md

@@ -1,8 +1,8 @@
-# Sahai
+# Saihai
 
 [English](README.md) | [Japanese](README.ja.md)
 
-Sahai is a local orchestrator and organization-runtime repository for running
+Saihai is a local orchestrator and organization-runtime repository for running
 AI-agent work through typed artifacts, explicit approval, durable state, and
 auditable evidence instead of treating a prompt as execution authority.
 
@@ -12,7 +12,7 @@ deterministic frontdoor, durable workflow runs, the constrained main-agent
 bridge, and typed report and evidence gates. Pre-release records may still use
 the ATV name for historical artifacts and compatibility aliases.
 
-Sahai uses only the Python 3.10+ standard library for normal operation. No
+Saihai uses only the Python 3.10+ standard library for normal operation. No
 `pip install` step is required.
 
 ## Release
@@ -85,7 +85,8 @@ Linked worktrees reuse the primary checkout's catalog.
 
 ```sh
 python3 scripts/setup_directory_paths.py --help
-# Supply all nine required directory options, then validate the catalog.
+# Supply the eight directory options shown in the configuration guide, then
+# validate the catalog. SAIHAI_ROOT defaults to the primary checkout.
 python3 scripts/setup_directory_paths.py --check
 ```
 
@@ -115,7 +116,7 @@ for the complete path audit.
 
 | Capability | Current behavior |
 |---|---|
-| Prompt classification | `scripts/configure_organization.py classify` and `/api/decide` classify work as `fast`, `strict`, or `maintenance`. Every mode still requires the applicable task and Vault records. |
+| Prompt classification | `scripts/configure_organization.py classify` and `/api/decide` report a `fast` or `strict` execution mode separately from the `enabled`, `disabled`, or `maintenance` organization state. Every mode and state still requires the applicable task and Vault records. |
 | Workflow selection | `workflow_selector.py` deterministically maps a typed classification to an active workflow template. A raw prompt is never selection authority. |
 | Frontdoor proposal | Prompt-originated requests stop at `proposed` or `waiting_human`, or fail closed as `blocked`. `propose` cannot produce an approved activation or create a workflow run. |
 | Approval | `approve` verifies a challenge derived from the proposal digest. Accepted activation sources are `human_ui`, `manual_cli`, and `orchestrator-start`, with trusted execution principals `human_operator`, `manual_operator`, and `orchestrator_start`. The narrow CLI defaults to `human_operator` / `human-ui` / `local_ui`. |
@@ -284,8 +285,8 @@ skills, automation, and the existing runtime.
 ```sh
 python3 scripts/configure_organization.py status
 python3 scripts/configure_organization.py runtime-paths
-python3 scripts/configure_organization.py classify --prompt "Review the latest forecast"
-AGENT_ORG_MAINTENANCE=1 python3 scripts/configure_organization.py classify --prompt "Repair a hook"
+python3 scripts/configure_organization.py classify --prompt "Review the latest forecast" --mode strict --organization-state enabled
+python3 scripts/configure_organization.py classify --prompt "Repair a hook" --organization-state maintenance
 python3 scripts/configure_organization.py validate-all
 python3 scripts/configure_organization.py workflow-selector validate-contracts
 python3 scripts/configure_organization.py workflow-frontdoor --help
@@ -295,7 +296,7 @@ python3 scripts/configure_organization.py workflow-frontdoor --help
 |---|---|
 | `status` | Print organization settings, role and policy counts, and repository root as JSON. |
 | `runtime-paths` | Verify the ITB runtime, workflow selector/frontdoor/server, operator CLI, and registry mirrors. |
-| `classify` | Classify a prompt as `fast`, `strict`, or `maintenance`. |
+| `classify` | Classify a prompt with optional `--mode {fast,strict}` and the separate `--organization-state {enabled,disabled,maintenance}` control. |
 | `validate-all` | Run the offline suites, contract validation, and Python compile check. |
 | `workflow-selector` | Validate workflow contracts and perform deterministic selection and activation-envelope operations. |
 | `workflow-frontdoor` | Provide the complete host-owned frontdoor and recovery surface. |
@@ -386,7 +387,7 @@ is added and attested.
 ### Scoped worker backend
 
 The live scoped-worker backend fails closed until a host operator manually
-configures the following assets. Sahai never generates keys or credentials.
+configures the following assets. Saihai never generates keys or credentials.
 
 The executor implementation is shipped, but the live `managed_worker` claim is
 suppressed. Codex 0.144.1 cannot prove absolute denial of a same-rootfs local
@@ -407,7 +408,7 @@ worker system.
 |---|---|
 | `SAIHAI_SCOPED_EXECUTOR_KEY_FILE` | Capability HMAC key in a regular, non-symlink, `0600` file containing at least 32 bytes. |
 | `SAIHAI_SCOPED_WORKTREE_ROOT` | Canonical root from which the host derives the task/run-bound worktree path. |
-| `SAIHAI_SCOPED_REPO_ROOT` | Host-owned absolute repository path. Defaults to the Sahai repository root. |
+| `SAIHAI_SCOPED_REPO_ROOT` | Host-owned absolute repository path. Defaults to the Saihai repository root. |
 | `SAIHAI_SCOPED_CODEX_EXECUTABLE` | Absolute pinned Codex CLI path whose digest is bound into the work order and capability. Group/world-writable binaries are rejected. |
 | `SAIHAI_SCOPED_CODEX_HOME` | Dedicated worker runtime/auth root; the main-agent profile is not inherited. |
 | `SAIHAI_ENABLE_SCOPED_WORKER_LIVE=1` | Explicit live-execution gate. Without it, only the deterministic fake harness is available. |
@@ -524,7 +525,7 @@ canonical root loaded from the primary checkout catalog.
 | `GET /api/org?session=<id>` | Team role state, active task, and busy count. |
 | `GET /api/role?session=<id>&role=<role_id>` | Role metadata, inbox, latest report, and provider evidence. |
 | `GET /api/config` | Organization settings and role/policy indexes. |
-| `GET /api/decide?prompt=<text>` | `fast`, `strict`, or `maintenance` classification. |
+| `GET /api/decide?prompt=<text>` | Decision with a `fast` or `strict` mode and a separate `enabled`, `disabled`, or `maintenance` organization state. |
 | `GET /api/workflow-runs?session=<id>&task=<id>&state=<state>` | Thin read-only workflow-run summaries. |
 | `GET /api/workflow-run?session=<id>&run=<id>` | Work order, report, provider evidence, and transition metadata. |
 | `GET /api/workflow-lock` | Global workflow-lock status for each configured orchestrator root. |

docs/configuration.md

@@ -28,6 +28,12 @@ audit is in
 
 Create the file in the primary checkout, or validate an existing file:
 
+The setup command requires the eight directory options shown below.
+`SAIHAI_ROOT`, the ninth required canonical path, defaults to the primary
+checkout that contains the catalog; pass `--saihai-root` only to override that
+default. Every supplied path must already exist as a readable directory, and
+the Agents Vault must also be writable.
+
 ```sh
 python3 scripts/setup_directory_paths.py \
   --agents-vault /absolute/path/to/Agents-Vault \
@@ -48,8 +54,8 @@ The setup command refuses to overwrite an existing file.
 
 The loader locates the catalog in this order:
 
-1. path supplied by process variable `SAIHAI_DIRECTORY_PATH_ENV` (or legacy `SAHAI_DIRECTORY_PATH_ENV`);
-2. `directory-path.env` below process `SAIHAI_ROOT` or legacy `SAHAI_ROOT`;
+1. path supplied by process variable `SAIHAI_DIRECTORY_PATH_ENV`;
+2. `directory-path.env` below process `SAIHAI_ROOT`;
 3. `directory-path.env` in the current Saihai checkout;
 4. `directory-path.env` in the primary checkout of a linked Git worktree;
 5. no file.
@@ -79,8 +85,10 @@ values.
 
 Normal Saihai entrypoints fail closed unless all nine canonical directory paths
 listed as required below exist and are readable. `AGENTS_VAULT_ROOT` must also
-be writable. Bootstrap-only consumers may load the catalog without enforcing
-the complete contract so that recovery remains possible.
+be writable. The setup command applies the same checks before creating the
+catalog and refuses to overwrite an existing file. Bootstrap-only consumers
+may load the catalog without enforcing the complete contract so that recovery
+remains possible.
 
 ## Canonical variables
 
@@ -111,6 +119,7 @@ emits value-free deprecation diagnostics.
 | `YASU_VAULT_ROOT` | `USER_VAULT_ROOT` |
 | `SKILLS_REPO_SKILLS_ROOT` | `SKILLS_ROOT` |
 | `DEV_REPO_ROOT` | `DEV_ROOT` |
+| `SAHAI_DIRECTORY_PATH_ENV` | `SAIHAI_DIRECTORY_PATH_ENV` |
 | `SAHAI_ORCH_STATE_ROOT` | `SAIHAI_ORCH_STATE_ROOT` |
 | `SAHAI_ITB_STATE_ROOTS` | `SAIHAI_ITB_STATE_ROOTS` |
 
@@ -132,5 +141,5 @@ python3 scripts/setup_directory_paths.py --check
 ```
 
 To recover from an invalid explicit selector, unset
-`SAIHAI_DIRECTORY_PATH_ENV` (and legacy `SAHAI_DIRECTORY_PATH_ENV`) so normal
-primary-checkout discovery can resume.
+`SAIHAI_DIRECTORY_PATH_ENV` so normal primary-checkout discovery can resume.
+Compatibility selector names are listed only in the table above.