feat: add governed chat agent selection

2026-07-08 22:19:51 +09:00
parent 06d8d31b16
commit 004afa73c9
18 changed files with 677 additions and 39 deletions
@@ -61,7 +61,7 @@
 | Future B1–B6 | 💤 vision | `CASAN_PLAN_FUTURE_PHASES.md` — approval workflow nâng cao · state machine · model benchmark · governed memory · auto-remediation · platform KPI. |
 | **17 Loop Engineering** | � T1–T6 done+test (offline) | **Agentic Loop Governance** — đủ 5 primitive + orchestrator (97/0 WSL, nối CI). T1 **Governor** (`loop-governor.py`; deny-by-default, no/corrupt policy→strict/HALT, on_exceed halt/escalate) 15/0; T2 **Convergence** (`loop-convergence.py`; repeat/thrash→OSCILLATING, flat→STALLED, fail-closed) 15/0; T3 **Verify Contract** (`loop-gate.py`; H4→DENY, unmet→FAIL, correction bounded→ESCALATE, no self-declared DONE) 20/0; T4 **Trace/Replay** (`loop-trace.py`; append-only hash-linked, edited→BREAK, tampered artifact→replay DRIFT) 16/0; T5 **Meta-loop** (`loop-metaloop.py`; propose≠apply, SoD, loosen>org_ceiling refused, apply qua governed CP store→đổi thật ceiling + rollback) 15/0; T6 **Orchestrator** (`loop-run.sh`; gate→governor→convergence→trace/turn, secure-by-default opt-out, nén giữa vòng) 16/0. State qua `CASAN_LOOP_STATE_ROOT` (repo `.specify/state` sạch). **Còn (infra):** T4 KMS-anchor head (A7 Vault), T6 widget Command Center (17.22, C5), live H3-judge. Chi tiết: `CASAN_PLAN_17_LOOP_ENGINEERING.md`. |
 | **16 Security audit remediation** | � P0/P1/P2 phần lớn done+test | **Remediation đã thực thi:** 28 SEC suite (151/0 WSL, nối `ci-harness-gate.sh`). Done: SEC-01..10, 12, 13, **14** (model-digest bỏ env-override ở prod/strict), 15, 16..21, **22** (trusted-time JWT `exp` ARCH-06 + tag proposal nguồn-không-tin ARCH-08), **26** (stored/second-order injection scan), 27..30, **23 Phase 1–5 offline** (multi-tenant: tenant-store+guard · per-tenant CP/audit/telemetry · RBAC data-boundary · tenant kill-switch/quota · ký registry · crypt at-rest per-tenant), **24 offline** (image digest-pin + ký workflow), **25 offline** (artifact attestation tested==deployed); **SEC-11 gộp vào SEC-17** (`CASAN_PROFILE=prod` enforce-by-default). **Còn 📋 planned (hạ tầng/process):** SEC-22 ARCH-10 (attestation ngoài) · SEC-23 23.11 (crypt qua Vault Transit) · **SEC-24 còn** (live CVE/OSV + scan image thật — offline image-pin/ký-workflow đã done) · **SEC-25 còn** (signed-commit enrollment + SLSA chain — offline artifact-attestation đã done). Chi tiết: `CASAN_PLAN_16` §0a/§2d. |
-| **18 Chat Console** | 🟡 **MVP-0 + MVP-1 done+test** · target arch còn planned | **Governed Chat Console** (cắt lát MVP chống lan man). **MVP-0 Ask CASAN read-only DONE**: `prompt-mode-router.py`, `chat-readonly.py`, `chat-session.schema.json`, H4 input/output scan, H5 chat audit hash-chain, H6 token telemetry, answer kèm evidence sources. **MVP-1 Operator DONE**: `operator-actions.yaml`, `chat-operator.py`, `chat-turn.py`, registered actions `run tests`/`build pack`/`verify pack`, all through `action-gate`, no free-command, action artifacts with provenance, Control Panel `POST /api/v1/chat/ask`, `GET /api/v1/chat/actions`, `/chat` UI. Test: chat suites **24/0** (`phase-chat-prompt-router` 8/0, `phase-chat-readonly` 5/0, `phase-chat-session-audit` 3/0, `phase-chat-operator` 8/0), Control Panel **24/0** + build xanh. Sau: MVP-2 chat-as-loop + agent/skill (**cần Plan-17+14**) → MVP-3 multi-tenant (**cần SEC-23**). |
+| **18 Chat Console** | 🟡 **MVP-0 + MVP-1 done+test; MVP-2 Track 4 foundation done** · target arch còn planned | **Governed Chat Console** (cắt lát MVP chống lan man). **MVP-0 Ask CASAN read-only DONE**: `prompt-mode-router.py`, `chat-readonly.py`, `chat-session.schema.json`, H4 input/output scan, H5 chat audit hash-chain, H6 token telemetry, answer kèm evidence sources. **MVP-1 Operator DONE**: `operator-actions.yaml`, `chat-operator.py`, `chat-turn.py`, registered actions `run tests`/`build pack`/`verify pack`, all through `action-gate`, no free-command, action artifacts with provenance. **MVP-2 Track 4 foundation DONE**: `agent-registry.yaml`, `chat-agent-resolver.py`, `chat:select_agent` RBAC, delegation hold, tool allowlist BLOCK, Control Panel `GET /api/v1/chat/agents` + `/chat` agent/skill/delegation picker. Test: chat suites **32/0** (`phase-chat-prompt-router` 8/0, `phase-chat-readonly` 5/0, `phase-chat-session-audit` 3/0, `phase-chat-operator` 8/0, `phase-chat-agent-select` 8/0), Control Panel **25/0** + build xanh. Next: MVP-2 Track 5/6 chat-as-loop + streaming + replay → MVP-3 multi-tenant. |
 ---

 ## Trần điểm & điều kiện lên "Strong (81+)"
@@ -35,7 +35,7 @@
 | 15 | `CASAN_PLAN_15_RESPONSIBLE_AI_DATA_GOV.md` | Responsible AI & Data Governance (FPT §14.3–14.4) | 📋 |
 | 16 | `CASAN_PLAN_16_SECURITY_AUDIT_REMEDIATION.md` | Security audit core harness + kế hoạch vá (tamper-evidence/injection/fail-open) | � P0/P1/P2 done |
 | 17 | `CASAN_PLAN_17_LOOP_ENGINEERING.md` | Loop Engineering / Agentic Loop Governance (budget governor, convergence, verify-contract, loop trace/replay, meta-loop) | 📋 |
-| 18 | `CASAN_PLAN_18_CHAT_CONSOLE.md` | Governed Chat Console (target arch; **MVP-0 Ask CASAN + MVP-1 Operator done** → chat-as-loop → multi-tenant) | 🟡 MVP-0/1 done |
+| 18 | `CASAN_PLAN_18_CHAT_CONSOLE.md` | Governed Chat Console (target arch; **MVP-0 Ask CASAN + MVP-1 Operator done; MVP-2 Track 4 foundation done** → chat-as-loop/streaming/replay → multi-tenant) | 🟡 MVP-0/1 + Track 4 done |
 | Future | `CASAN_PLAN_FUTURE_PHASES.md` | Approval workflow, state machine, benchmark, memory, remediation, KPI | 💤 vision |

 > **Không có plan số 11:** số 11 được bỏ trống có chủ ý — nhánh eval/traceability đã
@@ -1,12 +1,15 @@
 # KẾ HOẠCH 18 — Governed Chat Console (Chat-as-Loop qua Control Plane)

-> Status 2026-07-08: **🟡 MVP-0 + MVP-1 done+test — target architecture còn planned.**
+> Status 2026-07-08: **🟡 MVP-0 + MVP-1 done+test; MVP-2 Track 4 foundation done — target architecture còn planned.**
 > Đã implement **Ask CASAN — Read-only Evidence Assistant** qua harness + Control
 > Panel (`/api/v1/chat/ask`, `/chat` UI): Prompt Router deterministic
 > `READ_ONLY/OPERATOR/BLOCK/NOT_SUPPORTED`, context whitelist, H4 scan in/out, H5
 > chat audit hash-chain, H6 token telemetry, answer kèm evidence sources. **MVP-1
 > Operator** đã có registered actions `run tests` / `build pack` / `verify pack`
-> qua `action-gate`, không free-command, kết quả có artifact provenance. Mục tiêu
+> qua `action-gate`, không free-command, kết quả có artifact provenance. **MVP-2
+> Track 4 foundation** đã có `agent-registry.yaml`, agent/skill resolver, RBAC
+> gate chọn agent, delegation hold, tool allowlist block, Control Panel agent picker.
+> Mục tiêu
 > tổng thể vẫn là thêm **cửa sổ chat** vào Control Plane (Plan-13) như một
 > **bề mặt tương tác của core harness** — mỗi lượt chat là **một loop-run được
 > governance** (Plan-17), đi qua đúng H1→H7, không có đường vòng. Đây **không** phải
@@ -250,10 +253,10 @@ flowchart TD
 ### Track 4 — Agent/Skill selection governance `[MVP-2]`
 | Task | Việc | File | Verify (WSL) |
 |---|---|---|---|
-| 18.4.1 | `agent-registry.yaml` versioned (tool_allowlist, max_delegation, roles_allowed, model_role) + resolver bind theo turn | mới `config/agent-registry.yaml` | chọn agent → allowlist đúng |
-| 18.4.2 | RBAC gate chọn agent (14): role không đủ → DENY; vượt `max_delegation` → security-sensitive → approval | nối `rbac-check.py` | role thấp chọn agent tự chủ → DENY |
-| 18.4.3 | Tool-call runtime chỉ tool **trong allowlist**; ngoài → `action-gate` BLOCK | nối `action-gate.sh` | tool ngoài allowlist → BLOCK |
-| 18.4.4 | Skill code-gen chạy trong allowlist + H4 artifact-scan (Plan-02) trước output | nối `artifact-scan.sh` | skill sinh mã có injection → BLOCK |
+| 18.4.1 | ✅ `agent-registry.yaml` versioned (tool_allowlist, max_delegation, roles_allowed, model_role) + resolver bind theo turn | `packages/casan-harness/config/agent-registry.yaml` + `chat-agent-resolver.py` | chọn agent → allowlist đúng |
+| 18.4.2 | ✅ RBAC gate chọn agent (14): role không đủ → DENY; vượt `max_delegation` → security-sensitive → approval hold | nối `rbac-check.py` (`chat:select_agent`) | role thấp chọn agent tự chủ → DENY |
+| 18.4.3 | ✅ Tool-call runtime chỉ tool **trong allowlist**; ngoài → BLOCK trước runtime | `chat-agent-resolver.py` | tool ngoài allowlist → BLOCK |
+| 18.4.4 | ⏳ Skill code-gen chạy trong allowlist + H4 artifact-scan (Plan-02) trước output | nối `artifact-scan.sh` | skill sinh mã có injection → BLOCK |

 ### Track 5 — Chat-as-loop pipeline `[MVP-2]`
 | Task | Việc | File | Verify (WSL) |
@@ -274,7 +277,7 @@ flowchart TD
 |---|---|---|---|
 | 18.7.1 | ✅ Chat API (NestJS) **bọc** harness (MVP-0/1: read-only + operator; single-source, không verdict riêng) | `packages/casan-control-panel/backend/src/chat/*` | API gọi `chat-turn.py`; `console:test` xanh |
 | 18.7.2 | ✅ Chat panel React + hiển thị **evidence sources**, registered actions, action-gate + badge `mode/risk` | `packages/casan-control-panel/frontend/src/pages/Chat.tsx` | `console:build` xanh |
-| 18.7.3 | (MVP-2) agent/skill picker theo RBAC + badge `UNCERTIFIED/CERTIFIED` + nút **loop-breaker** + delegation-level | Control Plane UI | picker ẩn agent ngoài quyền |
+| 18.7.3 | 🟡 (MVP-2 partial) agent/skill picker theo RBAC + delegation-level đã có; còn badge `UNCERTIFIED/CERTIFIED` + nút **loop-breaker** khi Track 5/6 mở | Control Plane UI | picker hiện locked agent, backend DENY agent ngoài quyền |
 | 18.7.4 | Fail-loud: API/telemetry chết → UI `STALE/503` (tái dùng D3 `/healthz`) | nối Plan-07 D3 | ngắt backend → UI báo stale |

 ### Track 8 — Replay / Evidence / Command Center widgets `[MVP-2 → MVP-3]`
@@ -313,8 +316,8 @@ flowchart TD

 **File test (theo MVP):** MVP-0 — `phase-chat-prompt-router-tests.sh`,
 `phase-chat-readonly-tests.sh`, `phase-chat-session-audit-tests.sh`; MVP-1 —
-`phase-chat-operator-tests.sh`; MVP-2 — `phase-chat-pipeline-tests.sh`,
-`phase-chat-agent-select-tests.sh`, `phase-chat-stream-hold-tests.sh`,
+`phase-chat-operator-tests.sh`; MVP-2 — `phase-chat-agent-select-tests.sh`,
+`phase-chat-pipeline-tests.sh`, `phase-chat-stream-hold-tests.sh`,
 `phase-chat-replay-tests.sh`; MVP-3 — `phase-chat-tenant-tests.sh`. Chạy **WSL**
 (deterministic, mock model; nhánh live cần Ollama SKIP-aware). Nối `ci-harness-gate.sh`;
 cập nhật tổng test ở `CASAN_HARDENING_STATUS.md`.
@@ -336,6 +339,13 @@ cập nhật tổng test ở `CASAN_HARDENING_STATUS.md`.
 - [x] Kết quả action ghi artifact có provenance và được hiển thị lại trong `/chat`.
 - [x] Suite MVP-1 xanh và nối CI. Verify: `phase-chat-operator` 8/0, prompt-router 8/0, read-only 5/0, session-audit 3/0, `console:test` 24/0, `console:build` xanh.

+**MVP-2 Track 4 foundation (Agent/Skill selection governance):**
+- [x] Agent registry versioned có `tool_allowlist`, `max_delegation_level`, `roles_allowed`, `model_role`.
+- [x] Agent resolver bind theo turn, ghi audit hash-chain, fail-closed khi registry lỗi.
+- [x] RBAC gate `chat:select_agent`; role thấp chọn agent tự chủ → DENY; cross-tenant → DENY.
+- [x] Delegation vượt agent max → `REQUIRES_APPROVAL`; tool ngoài allowlist → BLOCK trước runtime.
+- [x] Control Panel `GET /api/v1/chat/agents`, `/chat` agent/skill/delegation picker, response có `agent_binding`. Verify: `phase-chat-agent-select` 8/0, RBAC 12/0, SEC-23 RBAC 6/0, `console:test` 25/0, `console:build` xanh.
+
 **Full (target architecture) — DoD tổng:**
 - [ ] Turn chạy **đúng như một loop-run Plan-17** (không định nghĩa vòng lặp riêng).
 - [ ] **Không đường vòng:** test chứng minh bỏ bất kỳ gate nào (preflight/verify) → FAIL.
@@ -349,9 +359,10 @@ cập nhật tổng test ở `CASAN_HARDENING_STATUS.md`.
 ---

 ## 7. Ghi chú trung thực & Non-goals (không lan man)
- **[MVP-0 + MVP-1] — 🟡 done+test.** Ask CASAN read-only và Operator registered
-  actions đã có harness CLI + Control Panel API/UI. Các phase Chat-as-loop /
-  Agent-selection / multi-tenant production vẫn chưa mở.
+- **[MVP-0 + MVP-1 + MVP-2 Track 4 foundation] — 🟡 done+test.** Ask CASAN read-only,
+  Operator registered actions, và agent/skill governance foundation đã có harness CLI
+  + Control Panel API/UI. Các phase Chat-as-loop / streaming / replay / multi-tenant
+  production vẫn chưa mở.
 - **MVP-first (chống lan man):** **MVP-0 (Ask CASAN read-only) KHÔNG phụ thuộc Plan-17/
  14/SEC-23** — làm được ngay trên nền H4/H5/H6 hiện có. Chỉ **MVP-2 trở đi** (chat-as-
  loop + agent) mới cần Plan-17 (T1–T3) + Plan-14 RBAC; **MVP-3** mới cần SEC-23 tenant.
@@ -77,7 +77,7 @@ approval JWT thật ([16 SEC-07](CASAN_PLAN_16_SECURITY_AUDIT_REMEDIATION.md)),
 |---|---|---|:--:|:--:|---|
 | B9 | ✅ **done** — **MVP-0 Ask CASAN read-only**: Prompt Router (Track 0) + Read-only Ask CASAN (Track 1) + session/audit (Track 2) + Control Panel API/UI (Track 7). H4 in/out, H5 audit hash-chain, H6 token, evidence sources; chat suites 14/0, Control Panel 23/0 + build xanh. Real model provider binding remains Track M follow-up when a provider is enabled. | [18 §1b, Track 0/1/2/7/M](CASAN_PLAN_18_CHAT_CONSOLE.md) | L | ✅ | done |
 | B10 | ✅ **done** — **MVP-1 Operator mode**: registered actions `run tests` / `build pack` / `verify pack`, no free-command, all through `action-gate`, action artifacts with provenance, Control Panel quick actions. `phase-chat-operator` 8/0; total chat suites 24/0; Control Panel 24/0 + build xanh. | [18 Track 3](CASAN_PLAN_18_CHAT_CONSOLE.md) | M | ✅ | done |
-| B11 | **MVP-2** Chat-as-loop (Track 5) + Agent/Skill selection (Track 4) + streaming draft-hold (Track 6) | [18 Track 4/5/6](CASAN_PLAN_18_CHAT_CONSOLE.md) | L | 🔗 | B6, C7(RBAC), B10 |
+| B11 | 🟡 **partial done** — Track 4 Agent/Skill governance foundation: `agent-registry.yaml`, `chat-agent-resolver.py`, RBAC `chat:select_agent`, delegation hold, tool allowlist BLOCK, Control Panel agent/skill/delegation picker. `phase-chat-agent-select` 8/0; Control Panel 25/0 + build xanh. **Remaining MVP-2:** Track 5 chat-as-loop + Track 6 streaming draft-hold + Track 8 replay widget. | [18 Track 4/5/6](CASAN_PLAN_18_CHAT_CONSOLE.md) | L | 🔗 | B6, C7(RBAC), B10 |
 | B12 | **Widget Loop/Chat** trên Command Center (loop ticker/budget/replay drawer) | [17 (17.22)](CASAN_PLAN_17_LOOP_ENGINEERING.md) · [18 Track 8](CASAN_PLAN_18_CHAT_CONSOLE.md) | M | 🔗 | C5 ✅ baseline |
 | B13 | **MVP-3** multi-tenant chat hardening (Track 9) | [18 Track 9](CASAN_PLAN_18_CHAT_CONSOLE.md) | M | 🔗 | A3(SEC-23) |

@@ -56,7 +56,7 @@ Approval inbox / HITL:
 - `POST /api/v1/approvals/decide` — approve/reject with SoD and reason; approved
  settings proposals apply through `control-plane-settings.py`.

-Governed Chat (Plan-18 MVP-0/1):
+Governed Chat (Plan-18 MVP-0/1 + MVP-2 Track 4 foundation):

 - `POST /api/v1/chat/ask` — Ask CASAN endpoint. The API only wraps harness
  `chat-turn.py`; router verdicts, H4 input/output scan, action-gate decisions,
@@ -64,11 +64,14 @@ Governed Chat (Plan-18 MVP-0/1):
  execution remain harness-owned.
 - `GET /api/v1/chat/actions` — list registered operator actions from
  `operator-actions.yaml`; no free-command execution is exposed.
+- `GET /api/v1/chat/agents` — list governed agents from `agent-registry.yaml`
+  with role visibility; selected agent/skill/delegation are bound by harness
+  `chat-agent-resolver.py`.
 - `GET /api/v1/chat/audit/verify` — verifies the chat audit hash chain.
 - `/chat` UI shows actor/role scope, `mode/risk/decision` badges, certified answer,
-  evidence sources, registered operator actions, action-gate status, router details,
-  and audit hash. Side-effect requests outside registered actions return governed
-  `BLOCK` or `NOT_SUPPORTED` responses.
+  evidence sources, registered operator actions, agent binding, action-gate status,
+  router details, and audit hash. Side-effect requests outside registered actions
+  return governed `BLOCK` or `NOT_SUPPORTED` responses.

 FinOps/SLO:

@@ -21,4 +21,9 @@ export class ChatController {
  actions(@Headers() headers: Record<string, string | string[] | undefined>) {
    return ok(this.svc.listActions(actorFromHeaders(headers)));
  }
+
+  @Get('agents')
+  agents(@Headers() headers: Record<string, string | string[] | undefined>) {
+    return ok(this.svc.listAgents(actorFromHeaders(headers)));
+  }
 }
@@ -7,6 +7,9 @@ import type { SettingsActor } from '../settings/settings.service.js';
 export interface ChatAskInput {
  message: string;
  chatId?: string;
+  agentId?: string;
+  skillId?: string;
+  delegationLevel?: number;
 }

 interface CommandResult {
@@ -18,6 +21,7 @@ interface CommandResult {
 const HARNESS_BIN = join(APP_ROOT, 'packages', 'casan-harness', 'scripts', 'bash');
 const CHAT_CLI = join(HARNESS_BIN, 'chat-turn.py');
 const OPERATOR_CLI = join(HARNESS_BIN, 'chat-operator.py');
+const AGENT_CLI = join(HARNESS_BIN, 'chat-agent-resolver.py');
 const RBAC_CLI = join(HARNESS_BIN, 'rbac-check.py');

 function runPython(script: string, args: string[]): CommandResult {
@@ -55,17 +59,25 @@ export class ChatService {
    }
    this.requireRead(actor);

-    const res = runPython(CHAT_CLI, [
+    const args = [
      'ask',
      '--message',
      input.message,
      '--actor',
      actor.actor,
+      '--role',
+      actor.role,
+      '--project',
+      actor.project,
      '--chat-id',
      input.chatId || 'chat-default',
      '--tenant',
      actor.tenant,
-    ]);
+    ];
+    if (input.agentId) args.push('--agent', input.agentId);
+    if (input.skillId) args.push('--skill', input.skillId);
+    if (input.delegationLevel !== undefined) args.push('--delegation-level', String(input.delegationLevel));
+    const res = runPython(CHAT_CLI, args);
    const parsed = parseJson<Record<string, any>>(res.stdout);
    if (parsed) {
      return { ...parsed, actor, audit_verify: this.verifyAudit() };
@@ -89,6 +101,14 @@ export class ChatService {
    throw new InternalServerErrorException(res.stderr || res.stdout || 'CHAT_OPERATOR_ACTIONS_FAILED');
  }

+  listAgents(actor: SettingsActor) {
+    this.requireRead(actor);
+    const res = runPython(AGENT_CLI, ['list-agents', '--role', actor.role]);
+    const parsed = parseJson<Record<string, any>>(res.stdout);
+    if (parsed) return parsed;
+    throw new InternalServerErrorException(res.stderr || res.stdout || 'CHAT_AGENTS_FAILED');
+  }
+
  private requireRead(actor: SettingsActor) {
    const res = runPython(RBAC_CLI, [
      'check',
@@ -6,6 +6,7 @@ import { join } from 'node:path';
 import { ChatService } from '../src/chat/chat.service.js';

 const viewer = { actor: 'chat-viewer', role: 'viewer', project: 'default', tenant: 'default' };
+const operator = { actor: 'chat-operator', role: 'operator', project: 'default', tenant: 'default' };

 function withTempChatState(fn: () => void) {
  const saved = {
@@ -66,15 +67,38 @@ test('chat ask marks side-effect requests unsupported in MVP-0', () => {
 test('chat ask executes registered operator action through action-gate', () => {
  withTempChatState(() => {
    const svc = new ChatService();
-    const actions = svc.listActions(viewer) as any;
+    const actions = svc.listActions(operator) as any;
    assert.ok(actions.actions.some((a: any) => a.id === 'run-chat-tests'));
+    const agents = svc.listAgents(operator) as any;
+    assert.ok(agents.agents.some((a: any) => a.id === 'ops-operator' && a.allowed_for_role === true));

-    const res = svc.ask({ message: 'run tests', chatId: 'operator-chat' }, viewer) as any;
+    const res = svc.ask({
+      message: 'run tests',
+      chatId: 'operator-chat',
+      agentId: 'ops-operator',
+      skillId: 'registered-actions',
+    }, operator) as any;
    assert.equal(res.success, true);
    assert.equal(res.mode, 'OPERATOR');
    assert.equal(res.decision, 'ACTION_COMPLETED');
    assert.equal(res.action.id, 'run-chat-tests');
    assert.equal(res.action_gate.outcome, 'ALLOW');
+    assert.equal(res.agent_binding.agent_selected, 'ops-operator');
+    assert.equal(res.agent_binding.skill_selected, 'registered-actions');
    assert.equal(res.audit_verify.ok, true);
  });
 });
+
+test('chat ask denies selected agent outside actor role', () => {
+  withTempChatState(() => {
+    const svc = new ChatService();
+    const agents = svc.listAgents(viewer) as any;
+    assert.ok(agents.agents.some((a: any) => a.id === 'ops-operator' && a.allowed_for_role === false));
+
+    const res = svc.ask({ message: 'run tests', agentId: 'ops-operator' }, viewer) as any;
+    assert.equal(res.success, false);
+    assert.equal(res.decision, 'DENIED');
+    assert.equal(res.agent_selected, 'ops-operator');
+    assert.equal(res.reason, 'role_not_allowed_for_agent');
+  });
+});
@@ -108,6 +108,19 @@ export interface ChatAnswer {
  };
  action?: { id: string; label: string; description: string } | null;
  action_gate?: { outcome?: string; reason?: string; exit_code?: number };
+  agent_binding?: {
+    success: boolean;
+    decision: string;
+    reason: string;
+    agent_selected: string;
+    skill_selected: string;
+    delegation_level: number;
+    model_role?: string;
+    mode?: string;
+    tool_allowlist: string[];
+    requires_approval: boolean;
+    audit?: { record_hash?: string; head?: string; seq?: number };
+  };
  actor: SettingsActor;
 }

@@ -118,6 +131,19 @@ export interface ChatAction {
  triggers: string[];
 }

+export interface ChatAgent {
+  id: string;
+  label: string;
+  mode: string;
+  model_role: string;
+  roles_allowed: string[];
+  skills_allowed: string[];
+  tool_allowlist: string[];
+  max_delegation_level: number;
+  requires_approval?: boolean;
+  allowed_for_role?: boolean;
+}
+
 export interface SettingsState {
  actor: SettingsActor;
  capabilities: {
@@ -164,10 +190,11 @@ export const api = {
    post<{ proposal: any; audit_verify: { ok: boolean; output: string } }>('approvals/submit', body, actorHeaders(actor)),
  decideApproval: (actor: SettingsActor, body: { id: string; decision: 'approve' | 'reject'; reason: string }) =>
    post<{ proposal: any; applied: any; audit_verify: { ok: boolean; output: string } }>('approvals/decide', body, actorHeaders(actor)),
-  askChat: (actor: SettingsActor, body: { message: string; chatId?: string }) =>
+  askChat: (actor: SettingsActor, body: { message: string; chatId?: string; agentId?: string; skillId?: string; delegationLevel?: number }) =>
    post<ChatAnswer>('chat/ask', body, actorHeaders(actor)),
  verifyChatAudit: () => get<{ ok: boolean; output: string }>('chat/audit/verify'),
  chatActions: (actor: SettingsActor) => getWithHeaders<{ success: boolean; actions: ChatAction[] }>('chat/actions', actorHeaders(actor)),
+  chatAgents: (actor: SettingsActor) => getWithHeaders<{ success: boolean; agents: ChatAgent[] }>('chat/agents', actorHeaders(actor)),
 };

 // Health is raw (not enveloped) + carries HTTP status.
@@ -1,6 +1,6 @@
 import { useState } from 'react';
 import { useMutation, useQuery } from '@tanstack/react-query';
-import { api, ChatAction, ChatAnswer, SettingsActor } from '../lib/api';
+import { api, ChatAction, ChatAgent, ChatAnswer, SettingsActor } from '../lib/api';
 import { Card, StatusBadge } from '../components/ui/Card';

 const ROLES = ['viewer', 'auditor', 'operator', 'project-admin', 'org-admin'];
@@ -26,6 +26,9 @@ export function Chat() {
    tenant: 'default',
  });
  const [chatId, setChatId] = useState('chat-default');
+  const [agentId, setAgentId] = useState('evidence-reader');
+  const [skillId, setSkillId] = useState('evidence-summary');
+  const [delegationLevel, setDelegationLevel] = useState(0);
  const [message, setMessage] = useState('Summarize Plan 18 MVP-0 status');
  const [last, setLast] = useState<ChatAnswer | null>(null);
  const [error, setError] = useState<string | null>(null);
@@ -40,9 +43,25 @@ export function Chat() {
    queryFn: () => api.chatActions(actor),
    retry: false,
  });
+  const agentsQuery = useQuery({
+    queryKey: ['chat-agents', actor],
+    queryFn: () => api.chatAgents(actor),
+    retry: false,
+  });
+  const agents = agentsQuery.data?.agents ?? [];
+  const selectedAgent = agents.find((a) => a.id === agentId) ?? agents.find((a) => a.allowed_for_role) ?? agents[0];
+  const selectedSkill = selectedAgent?.skills_allowed.includes(skillId)
+    ? skillId
+    : (selectedAgent?.skills_allowed[0] ?? '');

  const ask = useMutation({
-    mutationFn: (override?: { message?: string }) => api.askChat(actor, { message: override?.message ?? message, chatId }),
+    mutationFn: (override?: { message?: string; agentId?: string; skillId?: string }) => api.askChat(actor, {
+      message: override?.message ?? message,
+      chatId,
+      agentId: override?.agentId ?? selectedAgent?.id ?? agentId,
+      skillId: override?.skillId ?? selectedSkill,
+      delegationLevel,
+    }),
    onSuccess: (res) => {
      setLast(res);
      setError(null);
@@ -111,6 +130,33 @@ export function Chat() {
              <input className="w-full rounded border border-gray-300 px-3 py-2" value={chatId}
                onChange={(e) => setChatId(e.target.value)} />
            </label>
+            <label className="space-y-1 md:col-span-2 xl:col-span-1">
+              <span className="text-gray-500">Agent</span>
+              <select className="w-full rounded border border-gray-300 px-3 py-2" value={selectedAgent?.id ?? agentId}
+                onChange={(e) => {
+                  const next = agents.find((a) => a.id === e.target.value);
+                  setAgentId(e.target.value);
+                  setSkillId(next?.skills_allowed[0] ?? '');
+                }}>
+                {agents.map((a: ChatAgent) => (
+                  <option key={a.id} value={a.id}>
+                    {a.label}{a.allowed_for_role ? '' : ' (locked)'}
+                  </option>
+                ))}
+              </select>
+            </label>
+            <label className="space-y-1">
+              <span className="text-gray-500">Skill</span>
+              <select className="w-full rounded border border-gray-300 px-3 py-2" value={selectedSkill}
+                onChange={(e) => setSkillId(e.target.value)}>
+                {(selectedAgent?.skills_allowed ?? []).map((s) => <option key={s} value={s}>{s}</option>)}
+              </select>
+            </label>
+            <label className="space-y-1">
+              <span className="text-gray-500">Delegation</span>
+              <input className="w-full rounded border border-gray-300 px-3 py-2" type="number" min={0} max={5}
+                value={delegationLevel} onChange={(e) => setDelegationLevel(Number(e.target.value || 0))} />
+            </label>
          </div>
        </div>
      </Card>
@@ -125,6 +171,8 @@ export function Chat() {
              <StatusBadge value={last.mode} />
              <StatusBadge value={last.risk} />
              <StatusBadge value={last.decision} />
+              {last.agent_binding && <StatusBadge value={last.agent_binding.agent_selected} />}
+              {last.agent_binding && <StatusBadge value={`L${last.agent_binding.delegation_level}`} />}
              <StatusBadge value={last.audit_verify.ok ? 'audit ok' : 'audit fail'} />
            </div>
            <div className="whitespace-pre-wrap text-sm leading-6 text-gray-800">{last.answer}</div>
@@ -174,6 +222,18 @@ export function Chat() {
                  </div>
                </div>
              )}
+              {last.agent_binding && (
+                <div className="rounded border border-gray-200 p-3">
+                  <div className="text-xs font-semibold uppercase text-gray-400">Agent binding</div>
+                  <div className="mt-1 font-medium text-gray-800">{last.agent_binding.agent_selected}</div>
+                  <div className="mt-1 text-xs text-gray-500">{last.agent_binding.skill_selected || 'no skill'}</div>
+                  <div className="mt-2 flex flex-wrap gap-2">
+                    <StatusBadge value={last.agent_binding.decision} />
+                    <StatusBadge value={last.agent_binding.model_role ?? 'model'} />
+                    <StatusBadge value={`tools ${last.agent_binding.tool_allowlist.length}`} />
+                  </div>
+                </div>
+              )}
              <div>
                <div className="text-xs font-semibold uppercase text-gray-400">Matched rules</div>
                <div className="mt-1 flex flex-wrap gap-2">
@@ -202,7 +262,7 @@ export function Chat() {
                disabled={ask.isPending}
                onClick={() => {
                  setMessage(trigger);
-                  ask.mutate({ message: trigger });
+                  ask.mutate({ message: trigger, agentId: 'ops-operator', skillId: 'registered-actions' });
                }}
                className="text-left rounded border border-gray-200 p-3 hover:border-blue-300 hover:bg-blue-50 disabled:opacity-50"
              >
@@ -0,0 +1,36 @@
+{
+  "version": 1,
+  "agents": [
+    {
+      "id": "evidence-reader",
+      "label": "Evidence Reader",
+      "mode": "READ_ONLY",
+      "model_role": "read_only",
+      "roles_allowed": ["viewer", "auditor", "operator", "project-admin", "org-admin"],
+      "skills_allowed": ["evidence-summary"],
+      "tool_allowlist": [],
+      "max_delegation_level": 0
+    },
+    {
+      "id": "ops-operator",
+      "label": "Ops Operator",
+      "mode": "OPERATOR",
+      "model_role": "operator",
+      "roles_allowed": ["operator", "project-admin", "org-admin"],
+      "skills_allowed": ["registered-actions"],
+      "tool_allowlist": ["run-chat-tests", "build-evidence-pack", "verify-evidence-pack"],
+      "max_delegation_level": 1
+    },
+    {
+      "id": "codegen-draft",
+      "label": "Codegen Draft",
+      "mode": "CODEGEN",
+      "model_role": "codegen",
+      "roles_allowed": ["project-admin", "org-admin"],
+      "skills_allowed": ["sourcegen-draft"],
+      "tool_allowlist": ["artifact-scan", "tool-output-scan"],
+      "max_delegation_level": 2,
+      "requires_approval": true
+    }
+  ]
+}
@@ -0,0 +1,268 @@
+#!/usr/bin/env python3
+"""Plan-18 MVP-2 agent/skill governance resolver.
+
+This is a harness-owned policy primitive. It binds a chat turn to an agent,
+skill, delegation level, model role, and tool allowlist before any runtime is
+invoked. Unknown policy, RBAC failure, role mismatch, delegation escalation, or
+tool outside the agent allowlist all fail closed.
+"""
+import argparse
+import datetime
+import hashlib
+import json
+import os
+import subprocess
+import sys
+
+
+def project_root() -> str:
+    d = os.path.abspath(os.path.dirname(__file__))
+    p = d
+    while p != os.path.dirname(p):
+        if os.path.isdir(os.path.join(p, ".specify")) or os.path.isdir(os.path.join(p, "packages/casan-harness")):
+            return p
+        p = os.path.dirname(p)
+    return os.path.abspath(os.path.join(d, "..", "..", ".."))
+
+
+ROOT = project_root()
+HARNESS_ROOT = os.path.join(ROOT, "packages", "casan-harness")
+HARNESS_BIN = os.path.join(HARNESS_ROOT, "scripts", "bash")
+RBAC = os.path.join(HARNESS_BIN, "rbac-check.py")
+GENESIS_HASH = "0" * 64
+
+
+def now_iso() -> str:
+    return datetime.datetime.now(datetime.timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+
+
+def sha(text: str) -> str:
+    return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+def state_root() -> str:
+    return os.environ.get("CASAN_STATE_ROOT") or os.path.join(ROOT, ".specify")
+
+
+def registry_path() -> str:
+    return os.environ.get("CASAN_AGENT_REGISTRY_FILE") or os.path.join(HARNESS_ROOT, "config", "agent-registry.yaml")
+
+
+def audit_path() -> str:
+    return os.environ.get("CASAN_AGENT_BIND_AUDIT_LOG") or os.path.join(state_root(), "logs", "chat", "agent-bindings.jsonl")
+
+
+def head_path() -> str:
+    return os.environ.get("CASAN_AGENT_BIND_AUDIT_HEAD") or os.path.join(state_root(), "logs", "chat", "agent-bindings-head.txt")
+
+
+def load_registry():
+    try:
+        with open(registry_path(), encoding="utf-8") as fh:
+            data = json.load(fh)
+        agents = data.get("agents")
+        if not isinstance(agents, list):
+            raise ValueError("agents_not_list")
+        return data
+    except Exception as exc:
+        return {"_error": f"agent_registry_unreadable:{exc}", "agents": []}
+
+
+def find_agent(registry, agent_id: str):
+    return next((a for a in registry.get("agents", []) if a.get("id") == agent_id), None)
+
+
+def load_head() -> str:
+    try:
+        return open(head_path(), encoding="utf-8").read().strip() or GENESIS_HASH
+    except OSError:
+        return GENESIS_HASH
+
+
+def append_audit(base):
+    path = audit_path()
+    os.makedirs(os.path.dirname(path), exist_ok=True)
+    seq = 1
+    if os.path.isfile(path):
+        with open(path, encoding="utf-8") as fh:
+            seq = sum(1 for line in fh if line.strip()) + 1
+    prev = load_head()
+    core = {"seq": seq, **base, "prev_hash": prev}
+    record_hash = sha(json.dumps(core, sort_keys=True, ensure_ascii=False))
+    rec = {**core, "record_hash": record_hash}
+    with open(path, "a", encoding="utf-8") as fh:
+        fh.write(json.dumps(rec, ensure_ascii=False) + "\n")
+    os.makedirs(os.path.dirname(head_path()), exist_ok=True)
+    with open(head_path(), "w", encoding="utf-8") as fh:
+        fh.write(record_hash + "\n")
+    return rec
+
+
+def decision_payload(args, decision: str, reason: str, agent=None, skill="", tools=None):
+    tools = tools or []
+    success = decision == "BOUND"
+    payload = {
+        "success": success,
+        "decision": decision,
+        "reason": reason,
+        "agent_selected": (agent or {}).get("id") or args.agent,
+        "skill_selected": skill or "",
+        "delegation_level": args.delegation_level,
+        "model_role": (agent or {}).get("model_role"),
+        "mode": (agent or {}).get("mode"),
+        "tool_allowlist": (agent or {}).get("tool_allowlist", []),
+        "requested_tools": tools,
+        "side_effect_allowed": bool((agent or {}).get("tool_allowlist")),
+        "requires_approval": bool((agent or {}).get("requires_approval")) or decision == "REQUIRES_APPROVAL",
+    }
+    rec = append_audit({
+        "timestamp": now_iso(),
+        "harness": "H5-chat-agent-bind",
+        "actor": args.actor,
+        "role": args.role,
+        "project": args.project,
+        "tenant": args.tenant,
+        "agent_selected": payload["agent_selected"],
+        "skill_selected": payload["skill_selected"],
+        "delegation_level": args.delegation_level,
+        "decision": decision,
+        "reason": reason,
+        "tool_allowlist_ref": sha(json.dumps(payload["tool_allowlist"], sort_keys=True)),
+        "requested_tools_ref": sha(json.dumps(tools, sort_keys=True)),
+    })
+    payload["audit"] = {"seq": rec["seq"], "record_hash": rec["record_hash"], "head": rec["record_hash"]}
+    return payload
+
+
+def run_rbac(args):
+    r = subprocess.run([
+        "python3", RBAC, "check",
+        "--role", args.role,
+        "--resource", "chat",
+        "--action", "select_agent",
+        "--role-project", args.project,
+        "--target-project", args.target_project or args.project,
+        "--role-tenant", args.tenant,
+        "--target-tenant", args.target_tenant or args.tenant,
+    ], cwd=ROOT, capture_output=True, text=True)
+    return r.returncode, (r.stdout + r.stderr).strip()
+
+
+def bind(args) -> int:
+    registry = load_registry()
+    if registry.get("_error"):
+        out = decision_payload(args, "DENIED", registry["_error"])
+        print(json.dumps(out, ensure_ascii=False))
+        return 2
+
+    agent = find_agent(registry, args.agent)
+    if not agent:
+        out = decision_payload(args, "DENIED", "unknown_agent")
+        print(json.dumps(out, ensure_ascii=False))
+        return 2
+
+    rc, msg = run_rbac(args)
+    if rc != 0:
+        out = decision_payload(args, "DENIED", f"rbac_denied:{msg}", agent)
+        print(json.dumps(out, ensure_ascii=False))
+        return 2
+
+    if args.role not in agent.get("roles_allowed", []):
+        out = decision_payload(args, "DENIED", "role_not_allowed_for_agent", agent)
+        print(json.dumps(out, ensure_ascii=False))
+        return 2
+
+    skill = args.skill or (agent.get("skills_allowed", [""])[0] if agent.get("skills_allowed") else "")
+    if skill and skill not in agent.get("skills_allowed", []):
+        out = decision_payload(args, "DENIED", "skill_not_allowed_for_agent", agent, skill)
+        print(json.dumps(out, ensure_ascii=False))
+        return 2
+
+    max_level = int(agent.get("max_delegation_level", 0))
+    if args.delegation_level > max_level:
+        out = decision_payload(args, "REQUIRES_APPROVAL", "delegation_exceeds_agent_max", agent, skill)
+        print(json.dumps(out, ensure_ascii=False))
+        return 3
+
+    requested_tools = list(args.tool or [])
+    allow = set(agent.get("tool_allowlist", []))
+    outside = [t for t in requested_tools if t not in allow]
+    if outside:
+        out = decision_payload(args, "DENIED", "tool_not_allowlisted", agent, skill, requested_tools)
+        out["action_gate"] = {"outcome": "BLOCK", "reason": f"tool_not_allowlisted:{','.join(outside)}"}
+        print(json.dumps(out, ensure_ascii=False))
+        return 2
+
+    out = decision_payload(args, "BOUND", "agent_bound", agent, skill, requested_tools)
+    print(json.dumps(out, ensure_ascii=False))
+    return 0
+
+
+def list_agents(args) -> int:
+    registry = load_registry()
+    if registry.get("_error"):
+        print(json.dumps({"success": False, "error": registry["_error"], "agents": []}, ensure_ascii=False))
+        return 2
+    agents = []
+    for agent in registry.get("agents", []):
+        item = {k: agent.get(k) for k in (
+            "id", "label", "mode", "model_role", "roles_allowed", "skills_allowed",
+            "tool_allowlist", "max_delegation_level", "requires_approval",
+        )}
+        item["allowed_for_role"] = args.role in agent.get("roles_allowed", []) if args.role else None
+        agents.append(item)
+    print(json.dumps({"success": True, "version": registry.get("version"), "agents": agents}, ensure_ascii=False))
+    return 0
+
+
+def verify_audit() -> int:
+    prev = GENESIS_HASH
+    count = 0
+    try:
+        fh = open(audit_path(), encoding="utf-8")
+    except OSError:
+        print("CHAT_AGENT_AUDIT ok=true records=0 head=" + prev)
+        return 0
+    with fh:
+        for line in fh:
+            if not line.strip():
+                continue
+            count += 1
+            rec = json.loads(line)
+            got = rec.get("record_hash")
+            rest = {k: v for k, v in rec.items() if k != "record_hash"}
+            if rest.get("prev_hash") != prev or sha(json.dumps(rest, sort_keys=True, ensure_ascii=False)) != got:
+                print(f"CHAT_AGENT_AUDIT ok=false brokenAt={count}")
+                return 1
+            prev = got
+    print(f"CHAT_AGENT_AUDIT ok=true records={count} head={prev}")
+    return 0
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser()
+    sub = ap.add_subparsers(dest="cmd", required=True)
+    b = sub.add_parser("bind")
+    b.add_argument("--agent", required=True)
+    b.add_argument("--skill", default="")
+    b.add_argument("--actor", default="anonymous")
+    b.add_argument("--role", default="viewer")
+    b.add_argument("--project", default="default")
+    b.add_argument("--tenant", default="default")
+    b.add_argument("--target-project", default="")
+    b.add_argument("--target-tenant", default="")
+    b.add_argument("--delegation-level", type=int, default=0)
+    b.add_argument("--tool", action="append", default=[])
+    b.set_defaults(func=bind)
+
+    l = sub.add_parser("list-agents")
+    l.add_argument("--role", default="")
+    l.set_defaults(func=list_agents)
+
+    sub.add_parser("verify-audit").set_defaults(func=lambda _args: verify_audit())
+    args = ap.parse_args()
+    return args.func(args)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
@@ -24,6 +24,7 @@ def project_root() -> str:
 ROOT = project_root()
 BIN = os.path.join(ROOT, "packages", "casan-harness", "scripts", "bash")
 ROUTER = os.path.join(BIN, "prompt-mode-router.py")
+AGENT_RESOLVER = os.path.join(BIN, "chat-agent-resolver.py")
 READONLY = os.path.join(BIN, "chat-readonly.py")
 OPERATOR = os.path.join(BIN, "chat-operator.py")

@@ -36,13 +37,69 @@ def classify(message: str):
        return {"mode": "BLOCK", "risk": "high", "reason": "router_invalid_json", "matched_rules": [r.stderr.strip()]}


+def run_mode(args, binding):
+    r = subprocess.run(args, cwd=ROOT, capture_output=True, text=True)
+    try:
+        payload = json.loads(r.stdout)
+        payload["agent_binding"] = binding
+        print(json.dumps(payload, ensure_ascii=False))
+    except Exception:
+        if r.stdout:
+            sys.stdout.write(r.stdout)
+        if r.stderr:
+            sys.stderr.write(r.stderr)
+    return r.returncode
+
+
 def run_and_passthrough(args):
    r = subprocess.run(args, cwd=ROOT, text=True)
    return r.returncode


+def default_agent_for_mode(mode: str) -> str:
+    if mode == "OPERATOR":
+        return "ops-operator"
+    if mode == "CODEGEN":
+        return "codegen-draft"
+    return "evidence-reader"
+
+
+def bind_agent(args, router):
+    agent = args.agent or default_agent_for_mode(router.get("mode", "READ_ONLY"))
+    tools = []
+    if router.get("mode") == "OPERATOR":
+        # The operator primitive resolves the exact registered action later; the
+        # agent bind still proves this turn is allowed to use the operator class.
+        tools.append("run-chat-tests")
+    cmd = [
+        "python3", AGENT_RESOLVER, "bind",
+        "--agent", agent,
+        "--actor", args.actor,
+        "--role", args.role,
+        "--project", args.project,
+        "--tenant", args.tenant,
+        "--delegation-level", str(args.delegation_level),
+    ]
+    if args.skill:
+        cmd += ["--skill", args.skill]
+    for tool in tools:
+        cmd += ["--tool", tool]
+    r = subprocess.run(cmd, cwd=ROOT, capture_output=True, text=True)
+    if r.returncode != 0:
+        sys.stdout.write((r.stdout or r.stderr).strip() + "\n")
+        return r.returncode, None
+    try:
+        return 0, json.loads(r.stdout)
+    except Exception:
+        print(json.dumps({"success": False, "decision": "DENIED", "reason": "agent_resolver_invalid_json"}, ensure_ascii=False))
+        return 2, None
+
+
 def ask(args) -> int:
    router = classify(args.message)
+    bind_rc, binding = bind_agent(args, router)
+    if bind_rc != 0:
+        return bind_rc
    common = [
        "--message", args.message,
        "--actor", args.actor,
@@ -51,8 +108,8 @@ def ask(args) -> int:
        "--tenant", args.tenant,
    ]
    if router.get("mode") == "OPERATOR":
-        return run_and_passthrough(["python3", OPERATOR, "run", *common])
-    return run_and_passthrough(["python3", READONLY, "ask", *common])
+        return run_mode(["python3", OPERATOR, "run", *common], binding)
+    return run_mode(["python3", READONLY, "ask", *common], binding)


 def verify_audit() -> int:
@@ -65,9 +122,14 @@ def main() -> int:
    askp = sub.add_parser("ask")
    askp.add_argument("--message", required=True)
    askp.add_argument("--actor", default="anonymous")
+    askp.add_argument("--role", default="viewer")
+    askp.add_argument("--project", default="default")
    askp.add_argument("--chat-id", default="")
    askp.add_argument("--turn-id", default="")
    askp.add_argument("--tenant", default="default")
+    askp.add_argument("--agent", default="")
+    askp.add_argument("--skill", default="")
+    askp.add_argument("--delegation-level", type=int, default=0)
    askp.set_defaults(func=ask)
    sub.add_parser("verify-audit").set_defaults(func=lambda _args: verify_audit())
    args = ap.parse_args()
@@ -149,6 +149,7 @@ run "phase-chat-prompt-router"   bash "$TESTS/phase-chat-prompt-router-tests.sh"
 run "phase-chat-readonly"        bash "$TESTS/phase-chat-readonly-tests.sh"
 run "phase-chat-session-audit"   bash "$TESTS/phase-chat-session-audit-tests.sh"
 run "phase-chat-operator"        bash "$TESTS/phase-chat-operator-tests.sh"
+run "phase-chat-agent-select"    bash "$TESTS/phase-chat-agent-select-tests.sh"

 # ARCH-02: coverage cannot silently drop; ARCH-01: harness/policy cannot silently
 # drift. Both SKIP cleanly when no manifest is provisioned (non-strict dev/CI).
@@ -17,11 +17,11 @@ import sys
 # scope: "org" (all projects) or "project" (must match the acted-on project)
 PERMISSIONS = {
    "org-admin": {"scope": "org", "allow": {"*"}},
-    "project-admin": {"scope": "project", "allow": {"settings:read", "settings:write", "monitoring:read", "audit:read"}},
-    "approver": {"scope": "project", "allow": {"settings:read", "monitoring:read", "approval:grant"}},
-    "operator": {"scope": "project", "allow": {"monitoring:read", "kill_switch:engage"}},
-    "viewer": {"scope": "project", "allow": {"settings:read", "monitoring:read"}},
-    "auditor": {"scope": "org", "allow": {"settings:read", "monitoring:read", "audit:read"}},
+    "project-admin": {"scope": "project", "allow": {"settings:read", "settings:write", "monitoring:read", "audit:read", "chat:select_agent"}},
+    "approver": {"scope": "project", "allow": {"settings:read", "monitoring:read", "approval:grant", "chat:select_agent"}},
+    "operator": {"scope": "project", "allow": {"monitoring:read", "kill_switch:engage", "chat:select_agent"}},
+    "viewer": {"scope": "project", "allow": {"settings:read", "monitoring:read", "chat:select_agent"}},
+    "auditor": {"scope": "org", "allow": {"settings:read", "monitoring:read", "audit:read", "chat:select_agent"}},
 }

 # Maps an IdP-issued claim (role name / group) to an RBAC role. The IdP (Plan-07
@@ -0,0 +1,120 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# Plan-18 MVP-2 Track 4: agent/skill selection is governed by registry + RBAC.
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "$SCRIPT_DIR/../scripts/bash/casan-paths.sh"
+RESOLVER="$CASAN_HARNESS_ROOT/scripts/bash/chat-agent-resolver.py"
+TURN="$CASAN_HARNESS_ROOT/scripts/bash/chat-turn.py"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+export CASAN_STATE_ROOT="$WORK/state"
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+
+echo "===== Plan-18 MVP-2 agent/skill selection ====="
+
+python3 "$RESOLVER" list-agents --role viewer > "$WORK/list.json"
+python3 - "$WORK/list.json" <<'PY' \
+  && pass "agent registry lists versioned agents with role visibility" || fail "agent registry listing invalid"
+import json, sys
+d = json.load(open(sys.argv[1]))
+ids = {a["id"]: a for a in d["agents"]}
+assert d["success"] is True
+assert {"evidence-reader", "ops-operator", "codegen-draft"}.issubset(ids)
+assert ids["evidence-reader"]["allowed_for_role"] is True
+assert ids["ops-operator"]["allowed_for_role"] is False
+PY
+
+python3 "$RESOLVER" bind --agent evidence-reader --skill evidence-summary --role viewer --actor alice --project default --tenant tenant-a > "$WORK/viewer.json"
+python3 - "$WORK/viewer.json" <<'PY' \
+  && pass "viewer can bind read-only evidence-reader" || fail "viewer could not bind read-only agent"
+import json, sys
+d = json.load(open(sys.argv[1]))
+assert d["success"] is True
+assert d["decision"] == "BOUND"
+assert d["agent_selected"] == "evidence-reader"
+assert d["delegation_level"] == 0
+assert d["tool_allowlist"] == []
+PY
+
+set +e
+python3 "$RESOLVER" bind --agent ops-operator --role viewer --actor alice --project default --tenant tenant-a > "$WORK/viewer-op.json"
+RC=$?
+set -e 2>/dev/null || true
+python3 - "$WORK/viewer-op.json" "$RC" <<'PY' \
+  && pass "viewer cannot select operator agent" || fail "viewer selected operator agent"
+import json, sys
+d = json.load(open(sys.argv[1]))
+assert int(sys.argv[2]) == 2
+assert d["success"] is False
+assert d["decision"] == "DENIED"
+assert d["reason"] == "role_not_allowed_for_agent"
+PY
+
+set +e
+python3 "$RESOLVER" bind --agent evidence-reader --role viewer --actor alice --project default --tenant tenant-a --target-tenant tenant-b > "$WORK/cross-tenant.json"
+RC=$?
+set -e 2>/dev/null || true
+python3 - "$WORK/cross-tenant.json" "$RC" <<'PY' \
+  && pass "cross-tenant agent binding is denied by RBAC" || fail "cross-tenant agent binding was allowed"
+import json, sys
+d = json.load(open(sys.argv[1]))
+assert int(sys.argv[2]) == 2
+assert d["success"] is False
+assert d["decision"] == "DENIED"
+assert "rbac_denied" in d["reason"]
+PY
+
+set +e
+python3 "$RESOLVER" bind --agent evidence-reader --role viewer --actor alice --project default --tenant tenant-a --delegation-level 1 > "$WORK/delegation.json"
+RC=$?
+set -e 2>/dev/null || true
+python3 - "$WORK/delegation.json" "$RC" <<'PY' \
+  && pass "delegation above agent max requires approval" || fail "delegation escalation was not held"
+import json, sys
+d = json.load(open(sys.argv[1]))
+assert int(sys.argv[2]) == 3
+assert d["success"] is False
+assert d["decision"] == "REQUIRES_APPROVAL"
+assert d["requires_approval"] is True
+PY
+
+set +e
+python3 "$RESOLVER" bind --agent ops-operator --role operator --actor bob --project default --tenant tenant-a --tool shell > "$WORK/tool.json"
+RC=$?
+set -e 2>/dev/null || true
+python3 - "$WORK/tool.json" "$RC" <<'PY' \
+  && pass "tool outside agent allowlist is blocked" || fail "tool outside allowlist was accepted"
+import json, sys
+d = json.load(open(sys.argv[1]))
+assert int(sys.argv[2]) == 2
+assert d["success"] is False
+assert d["decision"] == "DENIED"
+assert d["action_gate"]["outcome"] == "BLOCK"
+PY
+
+set +e
+python3 "$TURN" ask --message "run tests" --agent ops-operator --role viewer --actor alice --chat-id agent-chat --tenant tenant-a > "$WORK/turn-deny.json"
+RC=$?
+set -e 2>/dev/null || true
+python3 - "$WORK/turn-deny.json" "$RC" <<'PY' \
+  && pass "chat-turn denies unauthorized selected agent before operator runtime" || fail "chat-turn bypassed agent resolver"
+import json, sys
+d = json.load(open(sys.argv[1]))
+assert int(sys.argv[2]) == 2
+assert d["success"] is False
+assert d["decision"] == "DENIED"
+assert d["agent_selected"] == "ops-operator"
+PY
+
+python3 "$RESOLVER" verify-audit > "$WORK/audit.out" \
+  && grep -q "CHAT_AGENT_AUDIT ok=true" "$WORK/audit.out" \
+  && pass "agent binding audit hash chain verifies" || fail "agent binding audit invalid"
+
+echo ""
+echo "===== CHAT AGENT SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -37,7 +37,7 @@ ids = {a["id"] for a in d["actions"]}
 assert {"run-chat-tests", "build-evidence-pack", "verify-evidence-pack"}.issubset(ids)
 PY

-python3 "$TURN" ask --message "run tests" --actor bob --chat-id op1 > "$WORK/run.json"
+python3 "$TURN" ask --message "run tests" --actor bob --role operator --chat-id op1 > "$WORK/run.json"
 python3 - "$WORK/run.json" <<'PY' \
  && pass "registered run tests action executes through chat-turn" || fail "registered action did not complete"
 import json, os, sys
@@ -1,7 +1,8 @@
 # CASAN Platform (Level 3 — Productization UI) · **PREVIEW**

 > Status: **PREVIEW.** The AgentOps dashboard and Plan-13 Control Panel, including
-> Command Center baseline and Plan-18 MVP-0/1 Chat Console, exist today.
+> Command Center baseline and Plan-18 MVP-0/1 Chat Console plus MVP-2 agent
+> selection foundation, exist today.
 > Evidence/attack viewers, Gitea integration, and managed production rollout remain
 > planned. `package-release.sh platform` builds a clearly-stamped
 > `casan-platform-preview-*` bundle containing only what exists.
@@ -13,12 +14,12 @@ Optional layer for teams that want UI / dashboard / visibility. Packages: `casan
 | Component | Status | Where |
 |---|---|---|
 | AgentOps Dashboard | ✅ exists | `packages/casan-harness/scripts/bash/dashboard-server.py` + `dashboard-serve.sh` (`casan dashboard`) |
-| **Ops Console (Control Panel)** | ✅ **monitoring + governed settings + HITL + kill-switch + FinOps/SLO + Command Center + Chat MVP-0/1 + local-prod TLS/OIDC smoke** | `packages/casan-control-panel/` — NestJS API + React UI over harness telemetry, settings management, approval inbox, kill-switch, FinOps/SLO, Command Center, Ask CASAN read-only, and registered Operator actions (`npm run console:api` + `console:ui`). Includes Run History + verdicts + governance + security + incidents + traceability + FinOps + Approvals + role-aware Settings page + `/command` evidence-backed executive view + `/chat` governed read-only/operator console. |
+| **Ops Console (Control Panel)** | ✅ **monitoring + governed settings + HITL + kill-switch + FinOps/SLO + Command Center + Chat MVP-0/1 + agent selection foundation + local-prod TLS/OIDC smoke** | `packages/casan-control-panel/` — NestJS API + React UI over harness telemetry, settings management, approval inbox, kill-switch, FinOps/SLO, Command Center, Ask CASAN read-only, registered Operator actions, and governed agent/skill selection (`npm run console:api` + `console:ui`). Includes Run History + verdicts + governance + security + incidents + traceability + FinOps + Approvals + role-aware Settings page + `/command` evidence-backed executive view + `/chat` governed read-only/operator console. |
 | Management / settings writes | ✅ done+test | Plan-13 Track 2: wraps `control-plane-settings.py`, calls `rbac-check.py`, supports set/rollback/audit verify. |
 | RBAC + approval inbox | ✅ local-prod done | Settings + kill-switch API RBAC enforcement, approval inbox/delegation/oversight, SoD, governed setting proposal apply, and local OIDC claim→role mapping smoke are done. Enterprise IdP rollout remains production follow-up. |
 | Evidence Pack Viewer | 📋 planned | reads `docs/output/casan/evidence-packs/` |
 | Attack Battery Viewer | 📋 planned | reads red-team corpus + H4 recall results |
-| Read-only Ask CASAN + Operator | ✅ MVP-0/1 done+test | Plan-18 MVP-0/1: `POST /api/v1/chat/ask`, `GET /api/v1/chat/actions` + `/chat`, backed by harness `chat-turn.py` |
+| Read-only Ask CASAN + Operator + agent selection | 🟡 MVP-0/1 done+test; MVP-2 Track 4 foundation done | Plan-18: `POST /api/v1/chat/ask`, `GET /api/v1/chat/actions`, `GET /api/v1/chat/agents` + `/chat`, backed by harness `chat-turn.py` and `chat-agent-resolver.py` |
 | Gitea webhook integration | 📋 planned | trigger gate / publish evidence on push |

 ## Build (preview)
@@ -28,6 +29,6 @@ scripts/package-release.sh platform     # → dist/casan-platform-preview-vX.Y.Z
 The bundle includes a `PREVIEW-INCOMPLETE.txt` marker. Do not treat it as a finished product.

 ## To implement later
-Start from Plan-15 RAI view or Plan-18 MVP-2 (Chat-as-loop + agent/skill). Keep new numbers
+Start from Plan-15 RAI view or Plan-18 MVP-2 Track 5/6 (Chat-as-loop + streaming). Keep new numbers
 evidence-backed with provenance; any write/governed action must continue to route through
 harness RBAC, approval, and audit primitives.