tycono 0.1.94-beta.0 → 0.1.94-beta.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tycono",
-  "version": "0.1.94-beta.0",
+  "version": "0.1.94-beta.2",
   "description": "Build an AI company. Watch them work.",
   "type": "module",
   "bin": {

package/src/api/src/engine/agent-loop.ts

@@ -172,7 +172,12 @@ export async function runAgentLoop(config: AgentConfig): Promise<AgentResult> {
   const hasBash = !readOnly && !!config.codeRoot;
   const node = orgTree.nodes.get(roleId);
   const heartbeatEnabled = node?.heartbeat?.enabled === true && subordinates.length > 0;
-  const tools = getToolsForRole(subordinates.length > 0, readOnly, hasBash, heartbeatEnabled);
+  // Peers = other roles with the same reportsTo (same parent in org tree)
+  const parentId = node?.reportsTo;
+  const hasPeers = parentId
+    ? (getSubordinates(orgTree, parentId).filter(id => id !== roleId).length > 0)
+    : false;
+  const tools = getToolsForRole(subordinates.length > 0, readOnly, hasBash, heartbeatEnabled, hasPeers);
 
   // 3. Set up tool executor
   const toolExecOptions: ToolExecutorOptions = {
@@ -550,163 +555,63 @@ export async function runAgentLoop(config: AgentConfig): Promise<AgentResult> {
       }
     }
 
-    // Phase B: Engineer/CTO Verification — type checking + visual verification
-    const verifiableRoles = ['engineer', 'cto'];
-    if (verifiableRoles.includes(roleId)) {
-      const hasFileChanges = allToolCalls.some((tc) =>
-        ['write', 'edit', 'bash'].includes(tc.name.toLowerCase()),
-      );
+    // Detect file changes once — used by Phase B and Post-K
+    const hasFileChanges = allToolCalls.some((tc) =>
+      ['write', 'edit', 'bash'].includes(tc.name.toLowerCase()),
+    );
 
-      if (hasFileChanges) {
-        const verifyPrompt = [
-          '[AUTO-VERIFICATION] 작업이 완료되었습니다. 아래 검증을 수행하세요:',
-          '1. `cd src/api && npx tsc --noEmit` — 타입 에러 확인',
-          '2. `cd src/web && npx tsc --noEmit` — 프론트엔드 타입 에러 확인',
-          '3. UI/CSS 변경이 있었다면 Playwright MCP로 스크린샷을 촬영하여 시각 검증',
-          '검증 결과를 간단히 보고하세요.',
-        ].join('\n');
-
-        messages.push({ role: 'user', content: verifyPrompt });
-
-        if (turns < maxTurns) {
-          turns++;
-          const verifyResponse = await llm.chat(context.systemPrompt, messages, tools, abortSignal);
-          totalInput += verifyResponse.usage.inputTokens;
-          totalOutput += verifyResponse.usage.outputTokens;
-          config.tokenLedger?.record({
-            ts: new Date().toISOString(),
-            sessionId: config.sessionId,
-            roleId,
-            model: config.model ?? 'unknown',
-            inputTokens: verifyResponse.usage.inputTokens,
-            outputTokens: verifyResponse.usage.outputTokens,
-          });
+    // Phase B: Member Self-Verification — type checking + visual verification
+    // Any non-C-Level role that made file changes gets verification (no hardcoded role IDs)
+    if (!isCLevel && hasFileChanges) {
+      const verifyPrompt = [
+        '[AUTO-VERIFICATION] 작업이 완료되었습니다. 아래 검증을 수행하세요:',
+        '1. `cd src/api && npx tsc --noEmit` — 타입 에러 확인',
+        '2. `cd src/web && npx tsc --noEmit` — 프론트엔드 타입 에러 확인',
+        '3. UI/CSS 변경이 있었다면 Playwright MCP로 스크린샷을 촬영하여 시각 검증',
+        '검증 결과를 간단히 보고하세요.',
+      ].join('\n');
 
-          messages.push({ role: 'assistant', content: verifyResponse.content });
-          for (const block of verifyResponse.content) {
-            if (block.type === 'text' && block.text) {
-              outputParts.push(block.text);
-              onText?.(block.text);
-            }
-          }
+      messages.push({ role: 'user', content: verifyPrompt });
 
-          // Execute verification tool calls if needed
-          if (verifyResponse.stopReason === 'tool_use') {
-            const verifyToolCalls = verifyResponse.content.filter(
-              (b): b is MessageContent & { type: 'tool_use' } => b.type === 'tool_use',
-            );
-            const verifyResults: ToolResult[] = [];
-            for (const tc of verifyToolCalls) {
-              allToolCalls.push({ name: tc.name, input: tc.input });
-              const result = await executeTool(
-                { id: tc.id, name: tc.name, input: tc.input },
-                toolExecOptions,
-              );
-              verifyResults.push(result);
-            }
-            messages.push({
-              role: 'user',
-              content: verifyResults.map((r) => ({
-                type: 'tool_result' as const,
-                tool_use_id: r.tool_use_id,
-                content: r.content,
-                is_error: r.is_error,
-              })) as unknown as MessageContent[],
-            });
+      if (turns < maxTurns) {
+        turns++;
+        const verifyResponse = await llm.chat(context.systemPrompt, messages, tools, abortSignal);
+        totalInput += verifyResponse.usage.inputTokens;
+        totalOutput += verifyResponse.usage.outputTokens;
+        config.tokenLedger?.record({
+          ts: new Date().toISOString(),
+          sessionId: config.sessionId,
+          roleId,
+          model: config.model ?? 'unknown',
+          inputTokens: verifyResponse.usage.inputTokens,
+          outputTokens: verifyResponse.usage.outputTokens,
+        });
 
-            if (turns < maxTurns) {
-              turns++;
-              const summaryResponse = await llm.chat(context.systemPrompt, messages, tools, abortSignal);
-              totalInput += summaryResponse.usage.inputTokens;
-              totalOutput += summaryResponse.usage.outputTokens;
-              config.tokenLedger?.record({
-                ts: new Date().toISOString(),
-                sessionId: config.sessionId,
-                roleId,
-                model: config.model ?? 'unknown',
-                inputTokens: summaryResponse.usage.inputTokens,
-                outputTokens: summaryResponse.usage.outputTokens,
-              });
-              for (const block of summaryResponse.content) {
-                if (block.type === 'text' && block.text) {
-                  outputParts.push(block.text);
-                  onText?.(block.text);
-                }
-              }
-            }
+        messages.push({ role: 'assistant', content: verifyResponse.content });
+        for (const block of verifyResponse.content) {
+          if (block.type === 'text' && block.text) {
+            outputParts.push(block.text);
+            onText?.(block.text);
           }
-
-          onTurnComplete?.(turns);
         }
 
-        // ── Post-Knowledging: ④⑤ 수행 프롬프트 (KP-008) ──
-        // Engineer/CTO가 구현 완료 후 The Loop 마무리 (Knowledge 업데이트 + Task 상태 갱신)를 수행하도록 유도
-        const postKPrompt = [
-          '[POST-KNOWLEDGING] 구현이 완료되었습니다. The Loop 마무리를 수행하세요:',
-          '',
-          '## ④ Knowledge 업데이트 (The Loop Step 4)',
-          '다음 중 해당하는 항목을 수행하세요:',
-          '- 본인 journal 업데이트 (`roles/' + roleId + '/journal/YYYY-MM-DD.md` — 오늘 날짜 파일)',
-          '- 구현 중 새로 발견한 패턴/아키텍처 결정이 있다면 관련 문서 업데이트',
-          '  (예: architecture/web-app-ia.md, architecture/session-worktree-isolation.md 등)',
-          '- 중요한 기술 결정은 operations/decisions/ 또는 architecture/ 반영',
-          '',
-          '## ⑤ Task 상태 갱신 (The Loop Step 5)',
-          '- `projects/tycono-platform/tasks.md` (또는 관련 tasks 파일)에서 완료한 태스크 상태를 DONE으로 변경',
-          '- 다음 작업이 있다면 식별하여 메모',
-          '',
-          '⛔ **필수**: ④와 ⑤를 모두 수행해야 The Loop이 완료됩니다.',
-          '이제 ④⑤를 수행하세요 (Read, Edit 도구 사용).',
-        ].join('\n');
-
-        messages.push({ role: 'user', content: postKPrompt });
-
-        // Run Post-K loop (최대 2턴 — C-Level의 3턴보다 짧게)
-        const maxPostKRounds = 2;
-        for (let round = 0; round < maxPostKRounds && turns < maxTurns; round++) {
-          if (abortSignal?.aborted) break;
-          turns++;
-
-          const postKResponse = await llm.chat(context.systemPrompt, messages, tools, abortSignal);
-          totalInput += postKResponse.usage.inputTokens;
-          totalOutput += postKResponse.usage.outputTokens;
-          config.tokenLedger?.record({
-            ts: new Date().toISOString(),
-            sessionId: config.sessionId,
-            roleId,
-            model: config.model ?? 'unknown',
-            inputTokens: postKResponse.usage.inputTokens,
-            outputTokens: postKResponse.usage.outputTokens,
-          });
-
-          messages.push({ role: 'assistant', content: postKResponse.content });
-          for (const block of postKResponse.content) {
-            if (block.type === 'text' && block.text) {
-              outputParts.push(block.text);
-              onText?.(block.text);
-            }
-          }
-
-          // If no tool calls, Post-K is done
-          if (postKResponse.stopReason !== 'tool_use') break;
-
-          // Execute Post-K tool calls
-          const postKToolCalls = postKResponse.content.filter(
+        // Execute verification tool calls if needed
+        if (verifyResponse.stopReason === 'tool_use') {
+          const verifyToolCalls = verifyResponse.content.filter(
             (b): b is MessageContent & { type: 'tool_use' } => b.type === 'tool_use',
           );
-          const postKResults: ToolResult[] = [];
-          for (const tc of postKToolCalls) {
+          const verifyResults: ToolResult[] = [];
+          for (const tc of verifyToolCalls) {
             allToolCalls.push({ name: tc.name, input: tc.input });
             const result = await executeTool(
               { id: tc.id, name: tc.name, input: tc.input },
               toolExecOptions,
             );
-            postKResults.push(result);
+            verifyResults.push(result);
           }
-
           messages.push({
             role: 'user',
-            content: postKResults.map((r) => ({
+            content: verifyResults.map((r) => ({
               type: 'tool_result' as const,
               tool_use_id: r.tool_use_id,
               content: r.content,
@@ -714,8 +619,110 @@ export async function runAgentLoop(config: AgentConfig): Promise<AgentResult> {
             })) as unknown as MessageContent[],
           });
 
-          onTurnComplete?.(turns);
+          if (turns < maxTurns) {
+            turns++;
+            const summaryResponse = await llm.chat(context.systemPrompt, messages, tools, abortSignal);
+            totalInput += summaryResponse.usage.inputTokens;
+            totalOutput += summaryResponse.usage.outputTokens;
+            config.tokenLedger?.record({
+              ts: new Date().toISOString(),
+              sessionId: config.sessionId,
+              roleId,
+              model: config.model ?? 'unknown',
+              inputTokens: summaryResponse.usage.inputTokens,
+              outputTokens: summaryResponse.usage.outputTokens,
+            });
+            for (const block of summaryResponse.content) {
+              if (block.type === 'text' && block.text) {
+                outputParts.push(block.text);
+                onText?.(block.text);
+              }
+            }
+          }
+        }
+
+        onTurnComplete?.(turns);
+      }
+    }
+
+    // ── Post-K: ④⑤ Knowledge/Task update (KP-008) ──
+    // ALL roles (C-Level and members) update journal/tasks after significant work.
+    // Runs for: members who made file changes, C-Level who dispatched work.
+    if (hasFileChanges || dispatches.length > 0) {
+      const postKPrompt = [
+        '[POST-KNOWLEDGING] 작업이 완료되었습니다. The Loop 마무리를 수행하세요:',
+        '',
+        '## ④ Knowledge 업데이트 (The Loop Step 4)',
+        '다음 중 해당하는 항목을 수행하세요:',
+        '- 본인 journal 업데이트 (`roles/' + roleId + '/journal/YYYY-MM-DD.md` — 오늘 날짜 파일)',
+        '- 구현 중 새로 발견한 패턴/아키텍처 결정이 있다면 관련 문서 업데이트',
+        '  (예: architecture/web-app-ia.md, architecture/session-worktree-isolation.md 등)',
+        '- 중요한 기술 결정은 operations/decisions/ 또는 architecture/ 반영',
+        '',
+        '## ⑤ Task 상태 갱신 (The Loop Step 5)',
+        '- `projects/tycono-platform/tasks.md` (또는 관련 tasks 파일)에서 완료한 태스크 상태를 DONE으로 변경',
+        '- 다음 작업이 있다면 식별하여 메모',
+        '',
+        '⛔ **필수**: ④와 ⑤를 모두 수행해야 The Loop이 완료됩니다.',
+        '이제 ④⑤를 수행하세요 (Read, Edit 도구 사용).',
+      ].join('\n');
+
+      messages.push({ role: 'user', content: postKPrompt });
+
+      // Run Post-K loop (최대 2턴)
+      const maxPostKRounds = 2;
+      for (let round = 0; round < maxPostKRounds && turns < maxTurns; round++) {
+        if (abortSignal?.aborted) break;
+        turns++;
+
+        const postKResponse = await llm.chat(context.systemPrompt, messages, tools, abortSignal);
+        totalInput += postKResponse.usage.inputTokens;
+        totalOutput += postKResponse.usage.outputTokens;
+        config.tokenLedger?.record({
+          ts: new Date().toISOString(),
+          sessionId: config.sessionId,
+          roleId,
+          model: config.model ?? 'unknown',
+          inputTokens: postKResponse.usage.inputTokens,
+          outputTokens: postKResponse.usage.outputTokens,
+        });
+
+        messages.push({ role: 'assistant', content: postKResponse.content });
+        for (const block of postKResponse.content) {
+          if (block.type === 'text' && block.text) {
+            outputParts.push(block.text);
+            onText?.(block.text);
+          }
+        }
+
+        // If no tool calls, Post-K is done
+        if (postKResponse.stopReason !== 'tool_use') break;
+
+        // Execute Post-K tool calls
+        const postKToolCalls = postKResponse.content.filter(
+          (b): b is MessageContent & { type: 'tool_use' } => b.type === 'tool_use',
+        );
+        const postKResults: ToolResult[] = [];
+        for (const tc of postKToolCalls) {
+          allToolCalls.push({ name: tc.name, input: tc.input });
+          const result = await executeTool(
+            { id: tc.id, name: tc.name, input: tc.input },
+            toolExecOptions,
+          );
+          postKResults.push(result);
         }
+
+        messages.push({
+          role: 'user',
+          content: postKResults.map((r) => ({
+            type: 'tool_result' as const,
+            tool_use_id: r.tool_use_id,
+            content: r.content,
+            is_error: r.is_error,
+          })) as unknown as MessageContent[],
+        });
+
+        onTurnComplete?.(turns);
       }
     }
   }

package/src/api/src/engine/context-assembler.ts

@@ -138,6 +138,11 @@ Use the code repository path for all source code work (reading, writing, buildin
     sections.push(buildSupervisionSection(node));
   }
 
+  // Knowledge consistency management (C-Level responsibility)
+  if (node.level === 'c-level') {
+    sections.push(buildKnowledgeManagementSection(roleId));
+  }
+
   // Dispatch 도구 안내 (하위 Role이 있는 경우)
   if (subordinates.length > 0) {
     sections.push(buildDispatchSection(orgTree, roleId, subordinates, options?.teamStatus));
@@ -526,36 +531,41 @@ ${subInfo}
 
 ## How to Dispatch
 
-**Dispatch is async: start a job → poll for result → review → next task.**
+**Dispatch is async: start a job → supervise with watch → review → next task.**
 
 \`\`\`bash
-# Step 1: Dispatch (returns immediately with job ID)
+# Step 1: Dispatch (returns immediately with session ID)
 python3 "$DISPATCH_CMD" ${exampleSubId} "Task description here"
+# → Session ID: ses-xxx
 
-# Step 2: Poll for result (repeat every 10-30s until DONE)
-python3 "$DISPATCH_CMD" --check <jobId>
+# Step 2: Supervise with watch (preferred) or poll with --check
+# Option A (preferred): Supervision watch — blocks server-side, returns digest
+python3 "$SUPERVISION_CMD" watch ses-xxx --duration 120
+# Option B (fallback): Poll for result
+python3 "$DISPATCH_CMD" --check ses-xxx
 \`\`\`
 
 ⛔ **NEVER use the Agent tool or Task tool to spawn sub-agents.** Those bypass the job tracking system. Use ONLY the dispatch command.
 
-### The Pattern: Dispatch → Poll → Review → Next
+### The Pattern: Dispatch → Watch → Analyze → Act
 
 \`\`\`bash
-# 1. Dispatch first task (returns job ID immediately)
+# 1. Dispatch tasks (save all session IDs)
 python3 "$DISPATCH_CMD" ${exampleSubId} "Task A"
-# Output includes: Job ID: job-xxx
+# → Session ID: ses-aaa
+${subordinates.length > 1 ? `python3 "$DISPATCH_CMD" ${subordinates[1]} "Task B"\n# → Session ID: ses-bbb` : ''}
 
-# 2. Poll for result (repeat until status is DONE)
-python3 "$DISPATCH_CMD" --check job-xxx
-# If RUNNING → wait 10-30s → --check again
-# If DONE → read the result → proceed to next task
+# 2. Enter supervision loop — watch ALL dispatched sessions
+python3 "$SUPERVISION_CMD" watch ses-aaa${subordinates.length > 1 ? ',ses-bbb' : ''} --duration 120
+# → Returns digest with significance score + activity summary
 
-# 3. Dispatch next task
-python3 "$DISPATCH_CMD" ${subordinates.length > 1 ? subordinates[1] : exampleSubId} "Task B"
-python3 "$DISPATCH_CMD" --check job-yyy
-# ... repeat
+# 3. Analyze digest → decide:
+#    - On track → watch again
+#    - Off track → amend: python3 "$SUPERVISION_CMD" amend ses-aaa "Fix: use adapter pattern"
+#    - Failed → abort: python3 "$SUPERVISION_CMD" abort ses-aaa --reason "wrong approach"
+#    - Need input → consult: python3 "$CONSULT_CMD" cbo "Is this direction right?"
 
-# 4. Continue until ALL tasks are done
+# 4. Repeat watch until ALL sessions complete
 \`\`\`
 
 ### --check Status Values
@@ -605,44 +615,52 @@ When you receive a directive:
 
 The loop:
 \`\`\`
-PLAN TASKS → DISPATCH → POLL (--check) → REVIEW RESULT → DECIDE
-                                                           ├── PASS → Next DIFFERENT Task
-                                                           └── FAIL → Re-dispatch with SPECIFIC fix
-                                                           └── ALL DONE → Update knowledge → Report
+PLAN TASKS → DISPATCH → SUPERVISION WATCH → ANALYZE DIGEST → DECIDE
+                                                               ├── On track → Watch again
+                                                               ├── Off track → Amend session
+                                                               ├── Failed → Abort + re-dispatch
+                                                               └── ALL DONE → Update knowledge → Report
 \`\`\`
 
 ### ⛔ CRITICAL: No Duplicate Dispatch
 
 **NEVER dispatch the same or similar task to the same role twice.**
-- If --check shows RUNNING, keep polling — do NOT re-dispatch
+- If watch shows sessions still running, keep watching — do NOT re-dispatch
 - If a subordinate completed a task, accept the result — do NOT re-dispatch
 - If the result is unsatisfactory, re-dispatch with SPECIFIC different instructions
-- Track dispatched job IDs — never repeat the same task
+- Track dispatched session IDs — never repeat the same task
 - After 2 dispatches to the same role, accept the result or report to CEO
 
 **Example: Full supervision session**
 \`\`\`bash
 # Task 1: Dispatch to engineer
 python3 "$DISPATCH_CMD" engineer "Implement feature X. Read tasks.md first."
-# → Job ID: job-001
+# → Session ID: ses-001
+
+# Supervision watch (blocks 120s, returns digest)
+python3 "$SUPERVISION_CMD" watch ses-001 --duration 120
+# → Digest: Engineer creating auth module (significance: 3, no anomalies)
+# Judgment: On track, continue watching
 
-# Poll until done
-python3 "$DISPATCH_CMD" --check job-001
-# → Status: RUNNING — check again in 10-30s
-python3 "$DISPATCH_CMD" --check job-001
-# → Status: DONE (result printed)
+python3 "$SUPERVISION_CMD" watch ses-001 --duration 120
+# → Digest: Engineer importing external API directly (significance: 7, anomaly: scope_creep)
+# Judgment: Wrong approach — amend!
+python3 "$SUPERVISION_CMD" amend ses-001 "Use our adapter pattern, not direct API import"
+
+python3 "$SUPERVISION_CMD" watch ses-001 --duration 120
+# → Digest: Engineer completed (msg:done)
 
 # Review result... looks good. Task 2 (QA):
 python3 "$DISPATCH_CMD" qa "Test feature X that engineer just implemented."
-# → Job ID: job-002
-python3 "$DISPATCH_CMD" --check job-002
-# → Status: DONE — found bugs
+# → Session ID: ses-002
+python3 "$SUPERVISION_CMD" watch ses-002 --duration 120
+# → Digest: QA found bugs (significance: 6)
 
 # Re-dispatch with SPECIFIC fix:
 python3 "$DISPATCH_CMD" engineer "Fix BUG: null check missing in auth.ts line 42"
-# → Job ID: job-003
-python3 "$DISPATCH_CMD" --check job-003
-# → Status: DONE — all good. Update knowledge and report.
+# → Session ID: ses-003
+python3 "$SUPERVISION_CMD" watch ses-003 --duration 120
+# → DONE — all good. Update knowledge and report.
 \`\`\`
 
 ⚠️ Do NOT use curl or other methods to create jobs — always use the dispatch command.
@@ -707,53 +725,110 @@ function buildSupervisionSection(node: OrgNode): string {
   const hb = node.heartbeat ?? { enabled: true, intervalSec: 120, maxTicks: 60 };
   return `# Supervision Mode (Heartbeat)
 
-⛔ **When you dispatch subordinates, you MUST enter supervision mode using heartbeat_watch.**
-⛔ **Do NOT use sleep+curl polling. heartbeat_watch blocks server-side at zero cost.**
+⛔ **After dispatching subordinates, you MUST enter supervision mode.**
+⛔ **Use \`python3 "$SUPERVISION_CMD" watch\` — it blocks server-side at zero LLM cost and returns a digest.**
+⛔ **Do NOT use sleep + \`--check\` polling. The supervision watch command is cheaper and gives better information.**
 
 ## Supervision Protocol
 
-1. **Dispatch** subordinates with clear task descriptions
-2. **Call heartbeat_watch** with the returned session IDs:
-   \`heartbeat_watch(sessionIds=[...], durationSec=${hb.intervalSec})\`
-3. **Analyze the digest** against your plan:
-   - On track → call heartbeat_watch again (keep watching)
-   - Off track → \`amend_session(sessionId, instruction)\` to course-correct
-   - Seriously wrong → \`abort_session(sessionId)\` + re-dispatch with different instructions
-   - Need peer input → \`consult(peer_role_id, question)\`
-   - All done → compile results and report to your superior
-4. **Repeat** heartbeat_watch until all subordinates complete
-
-## Available Supervision Tools
-
-| Tool | When to Use |
-|------|-------------|
-| \`heartbeat_watch\` | Watch subordinate sessions (blocks ${hb.intervalSec}s, $0 LLM cost) |
-| \`amend_session\` | Inject new instructions into a running session |
-| \`abort_session\` | Kill a session that's going wrong |
-| \`consult\` | Ask a peer C-Level for their perspective |
+1. **Dispatch** subordinates with clear task descriptions. Save returned session IDs.
+2. **Watch** with the supervision command:
+   \`\`\`bash
+   python3 "$SUPERVISION_CMD" watch <ses-id1>,<ses-id2> --duration ${hb.intervalSec}
+   \`\`\`
+   This blocks for ${hb.intervalSec}s (or until an alert event) and returns a digest summary.
+3. **Analyze the digest** — each tick, you must think critically:
+   - ✅ **Direction check**: Are subordinates on track with the plan?
+   - ✅ **Quality check**: Is the approach correct? (e.g., right patterns, right tools)
+   - ✅ **Peer consult**: Unsure about business/market direction? → \`python3 "$CONSULT_CMD" cbo "question"\`
+   - ⚠️ **Course correct**: Wrong direction → \`python3 "$SUPERVISION_CMD" amend <ses-id> "new instruction"\`
+   - 🛑 **Abort**: Seriously wrong → \`python3 "$SUPERVISION_CMD" abort <ses-id> --reason "why"\`
+   - ✅ **All done**: Compile results and report to your superior
+4. **Repeat** watch until all subordinates complete. Do NOT stop after one tick.
+
+## Supervision Commands
+
+| Command | Description |
+|---------|-------------|
+| \`python3 "$SUPERVISION_CMD" watch <ids> --duration ${hb.intervalSec}\` | Watch sessions (blocks ${hb.intervalSec}s, returns digest) |
+| \`python3 "$SUPERVISION_CMD" amend <ses-id> "instruction"\` | Inject new instructions into running session |
+| \`python3 "$SUPERVISION_CMD" abort <ses-id> --reason "why"\` | Kill a session going wrong |
+| \`python3 "$SUPERVISION_CMD" peers --wave <waveId> --role <roleId>\` | Discover peer C-Level sessions |
+| \`python3 "$CONSULT_CMD" <peer-role> "question"\` | Ask a peer C-Level for input |
 
 ## Digest Response
 
-heartbeat_watch returns a digest with:
-- **Significance score** (0-10): How much attention this tick needs
-- **Anomalies**: Errors, stalls (3min+), sessions awaiting input
-- **Per-session activity**: What each subordinate has been doing
-- **Peer activity** (if peers are also in supervision mode)
+The watch command returns a JSON digest with:
+- **significanceScore** (0-10): How much attention this tick needs
+- **anomalies**: Errors, stalls (3min+ no events), sessions awaiting input
+- **text**: Human-readable summary of per-session activity
+
+Score 0-1 = quiet (all normal). Score 5+ = needs attention. Score 8+ = likely needs intervention.
+
+## Tick Quality (CRITICAL)
 
-Quiet ticks (score 0-1) return a single line: "All N sessions progressing normally."
+⛔ **Each tick, you must make a REAL judgment — not just "still running, continuing".**
+
+Good tick response:
+\`\`\`
+Digest: Engineer working on auth module (seq 45), QA waiting.
+Judgment: Engineer is importing external API directly — should use our adapter pattern.
+Action: amend engineer session with instruction to use adapter.
+\`\`\`
+
+Bad tick response (DO NOT DO THIS):
+\`\`\`
+"Sessions are still running. Continuing to watch."
+\`\`\`
 
 ## Budget
 
 - Max ticks: ${hb.maxTicks} (${Math.round(hb.maxTicks * hb.intervalSec / 60)} minutes total)
-- Quiet tick cost: ~$0.001 (minimal LLM analysis)
-- Alert tick cost: ~$0.02-0.05 (intervention decision)
+- Quiet tick cost: ~$0.001 | Alert tick cost: ~$0.02-0.05
 
 ## ⛔ Anti-Patterns
 
-- ❌ Using \`bash_execute\` with sleep/curl to poll — use heartbeat_watch instead
-- ❌ Calling \`--check\` in a loop — heartbeat_watch handles this automatically
+- ❌ Using \`sleep\` + \`--check\` to poll — use supervision watch instead
+- ❌ Saying "still RUNNING, continuing to wait" without real analysis
 - ❌ Ignoring digest anomalies — always address errors and stalls
-- ❌ Not re-watching after a quiet tick — keep the loop going until all done`;
+- ❌ Stopping supervision after one tick — keep watching until ALL done
+- ❌ Not consulting peers when business/market judgment is needed`;
+}
+
+function buildKnowledgeManagementSection(roleId: string): string {
+  const domainScope = roleId === 'cto'
+    ? '`architecture/`, `knowledge/` (technical docs)'
+    : roleId === 'cbo'
+      ? '`knowledge/`, `company/` (business docs)'
+      : '`knowledge/` (your domain docs)';
+
+  return `# Knowledge Consistency Management (C-Level Responsibility)
+
+⛔ **You are responsible for keeping your domain's knowledge current and consistent.**
+
+## When to Check
+
+- After completing any task (The Loop step ④)
+- During supervision ticks when subordinates are progressing normally
+- When you notice conflicting information during work
+
+## Knowledge Gate
+
+1. Check if changes conflict with existing documents (grep 3+ keywords)
+2. Update or deprecate stale references — do NOT leave legacy docs pretending to be current
+3. Verify new docs are registered in their Hub
+4. Ensure cross-links between related documents
+
+## AKB Linting
+
+| Check | Description |
+|-------|-------------|
+| Orphan docs | Every document reachable from a Hub? |
+| Stale content | Deprecated designs still described as current? Mark ⚠️ |
+| Term consistency | Same concept called different names across docs? |
+| Date freshness | Old dates/status without warning? |
+
+**Your scope**: ${domainScope}`;
 }
 
 function buildConsultSection(orgTree: OrgTree, roleId: string): string | null {

package/src/api/src/engine/org-tree.ts

@@ -30,7 +30,7 @@ export interface HeartbeatConfig {
 export interface OrgNode {
   id: string;
   name: string;
-  level: 'c-level' | 'team-lead' | 'member';
+  level: 'c-level' | 'member';
   reportsTo: string;
   children: string[];
   persona: string;

package/src/api/src/engine/role-lifecycle.ts

@@ -19,7 +19,7 @@ export interface SkillExportDef {
 export interface RoleDefinition {
   id: string;
   name: string;
-  level: 'c-level' | 'team-lead' | 'member';
+  level: 'c-level' | 'member';
   reportsTo: string;
   persona: string;
   skills?: string[];