tycono 0.1.94-beta.6 → 0.1.94-beta.8

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tycono",
-  "version": "0.1.94-beta.6",
+  "version": "0.1.94-beta.8",
   "description": "Build an AI company. Watch them work.",
   "type": "module",
   "bin": {

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

@@ -745,11 +745,12 @@ function buildSupervisionSection(node: OrgNode): string {
    - 🛑 **Abort**: Seriously wrong → \`python3 "$SUPERVISION_CMD" abort <ses-id> --reason "why"\`
    - ✅ **All done?** → Before reporting done, **verify deliverables** (see Quality Gate below)
 4. **Repeat** watch until all subordinates complete. Do NOT stop after one tick.
-5. **Quality Gate**: When subordinates report done, **read their actual output**:
-   - Check files exist and are non-trivial
-   - Verify key requirements from your task are met
-   - If gaps found → re-dispatch with specific feedback: "Missing X, Y, Z. Continue."
-   - There is NO time limit. Iterate until the work truly meets the requirements.
+5. **Quality Gate**: When subordinates report done, **run and test** the output:
+   - For web apps/games: start a local server and open in browser to verify it actually works
+   - Try the core user interactions — if basic things don't work, it's NOT done
+   - Check that required libraries/tools mentioned in the task are actually used
+   - If gaps found → re-dispatch with **specific, actionable** feedback (not "improve quality")
+   - There is NO time limit. Non-working code is worse than less code that works.
 
 ## Supervision Commands
 

package/src/api/src/services/session-store.ts

@@ -263,6 +263,11 @@ export function deleteSession(id: string, force = false): boolean {
     console.warn(`[SessionStore] BLOCKED deletion of wave session ${id} (waveId=${session.waveId}, roleId=${session.roleId}). Use force=true to override.`);
     return false;
   }
+  // BUG-008 hard guard: CEO supervisor session is NEVER deletable during wave
+  if (session.roleId === 'ceo' && session.waveId && session.source === 'wave') {
+    console.error(`[SessionStore] HARD BLOCK: CEO supervisor session ${id} cannot be deleted (waveId=${session.waveId}). This is a 1:1:1 invariant.`);
+    return false;
+  }
 
   console.log(`[SessionStore] Deleting session ${id} (roleId=${session.roleId}, waveId=${session.waveId ?? 'none'}, messages=${session.messages.length})`);
   cache.delete(id);

package/src/api/src/services/supervisor-heartbeat.ts

@@ -296,16 +296,23 @@ ${recoveryContext}
 
 ## Quality Gate (CRITICAL — G-09)
 ⛔ **"Subordinate said done" ≠ "Work is actually done."**
-Before declaring yourself done, you MUST verify the deliverables meet the directive's requirements:
+⛔ **"Code exists" ≠ "Code works."** You MUST run and test the output, not just read files.
 
-1. **Read the actual output files** — don't trust status reports. Check the code, docs, or artifacts yourself.
-2. **Test if it works** — if the directive asks for a working game/app, check if it actually runs.
+Before declaring yourself done, you MUST:
+
+1. **Read the actual output files** — don't trust status reports. Check the code yourself.
+2. **RUN it and test it** — this is the most important step:
+   - For web apps/games: \`cd <code-dir> && python3 -m http.server 9999\` then open in browser
+   - Actually try the core interactions (click buttons, press keys, navigate)
+   - If basic interactions fail (can't move, can't click, blank screen) → it's NOT done
 3. **Count against requirements** — if the directive says "15 monsters, 7 maps", count them.
-4. **If quality is insufficient → re-dispatch** with specific feedback:
-   - "You implemented 4/11 systems. Still missing: NPC dialogue, inventory, capture. Continue."
-   - "The game doesn't load in browser. Fix the entry point and test."
-5. **Iterate until the directive is truly fulfilled.** There is NO time limit.
-   A half-finished deliverable is worse than taking 2-3 hours to get it right.
+4. **Check the directive's specific tech requirements** — if it mentions a specific library/engine, verify it's actually used in the code (grep for it).
+5. **If quality is insufficient → re-dispatch** with specific, actionable feedback:
+   - "Arrow keys don't move the player. Fix input handling in WorldScene."
+   - "TyconoForge was required but not used. Add character rendering with TyconoForge.render()."
+   - NOT vague feedback like "improve quality" or "make it better"
+6. **Iterate until the directive is truly fulfilled.** There is NO time limit.
+   20,000 lines of non-working code is worse than 5,000 lines that actually play.
 
 Re-dispatch pattern:
 - dispatch same C-Level with specific gaps identified
@@ -321,27 +328,33 @@ Re-dispatch pattern:
 6. If gaps exist → re-dispatch with specific feedback. Repeat 3-5.
 7. Only when ALL requirements are met → compile results and report`;
 
-    // Create supervisor session
-    const session = createSession('ceo', {
-      mode: 'do',
-      source: 'wave',
-      waveId: state.waveId,
-    });
-
-    state.supervisorSessionId = session.id;
+    // BUG-008 fix: Wave:Supervisor:Session = 1:1:1 invariant.
+    // Reuse existing session on restart instead of creating a new one.
+    let sessionId = state.supervisorSessionId;
+    if (sessionId && getSession(sessionId)) {
+      console.log(`[Supervisor] Reusing existing session ${sessionId} for wave ${state.waveId}`);
+    } else {
+      const session = createSession('ceo', {
+        mode: 'do',
+        source: 'wave',
+        waveId: state.waveId,
+      });
+      sessionId = session.id;
+      state.supervisorSessionId = sessionId;
+
+      // Add the directive as CEO message so the session isn't empty (prevents deleteEmpty cleanup)
+      const ceoMsg: Message = {
+        id: `msg-${Date.now()}-ceo-supervisor`,
+        from: 'ceo',
+        content: state.directive,
+        type: 'directive',
+        status: 'done',
+        timestamp: new Date().toISOString(),
+      };
+      addMessage(sessionId, ceoMsg);
+    }
     state.status = 'running';
 
-    // Add the directive as CEO message so the session isn't empty (prevents deleteEmpty cleanup)
-    const ceoMsg: Message = {
-      id: `msg-${Date.now()}-ceo-supervisor`,
-      from: 'ceo',
-      content: state.directive,
-      type: 'directive',
-      status: 'done',
-      timestamp: new Date().toISOString(),
-    };
-    addMessage(session.id, ceoMsg);
-
     try {
       const exec = executionManager.startExecution({
         type: 'wave',
@@ -349,14 +362,14 @@ Re-dispatch pattern:
         task: supervisorTask,
         sourceRole: 'ceo',
         targetRoles: state.targetRoles,
-        sessionId: session.id,
+        sessionId,
       });
 
       state.executionId = exec.id;
 
       this.watchExecution(state, exec);
 
-      console.log(`[Supervisor] Started for wave ${state.waveId} | session=${session.id} | exec=${exec.id}`);
+      console.log(`[Supervisor] Started for wave ${state.waveId} | session=${sessionId} | exec=${exec.id}`);
     } catch (err) {
       console.error(`[Supervisor] Failed to start for wave ${state.waveId}:`, err);
       state.status = 'error';