feat: show MEMORY.md raw content in Plan tab during interview/review

During interviewing/review status, the Plan tab now displays the raw
MEMORY.md content with simple markdown rendering instead of the
structured phase/step list. This lets users verify their input is being
captured in real-time. Switches back to structured view on executing.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

cmd/picoclaw/cmd_gateway.go

@@ -306,7 +306,7 @@ func (p *agentLoopDataProvider) ListSkills() []skills.SkillInfo {
 }
 
 func (p *agentLoopDataProvider) GetPlanInfo() miniapp.PlanInfo {
-	hasPlan, status, currentPhase, totalPhases, display := p.loop.GetPlanInfo()
+	hasPlan, status, currentPhase, totalPhases, display, memory := p.loop.GetPlanInfo()
 
 	// Convert agent.PlanPhase → miniapp.PlanPhase
 	agentPhases := p.loop.GetPlanPhases()
@@ -334,6 +334,7 @@ func (p *agentLoopDataProvider) GetPlanInfo() miniapp.PlanInfo {
 		TotalPhases:  totalPhases,
 		Display:      display,
 		Phases:       phases,
+		Memory:       memory,
 	}
 }
 

pkg/agent/loop.go

@@ -960,6 +960,13 @@ func buildTaskReminder(userMessage string, lastBlocker string) providers.Message
 	}
 }
 
+// interviewRejectMessage is the fixed rejection text injected when tool calls
+// are blocked during the interview phase.  It is deliberately short to avoid
+// wasting tokens, and ends with a purpose reminder to steer the LLM back.
+const interviewRejectMessage = "[System] Tool call rejected. " +
+	"You are in interview mode — ask the user questions and update MEMORY.md. " +
+	"Do not execute, edit, or write project files."
+
 // buildPlanReminder returns a reminder message for plan pre-execution states
 // (interviewing / review) to keep the AI focused on the interview workflow
 // during tool-call iterations.
@@ -1594,6 +1601,39 @@ func (al *AgentLoop) runLLMIteration(
 			normalizedToolCalls = append(normalizedToolCalls, providers.NormalizeToolCall(tc))
 		}
 
+		// --- Interview mode: reject disallowed tool calls before they
+		// enter messages or session history.  Rejected calls are stripped
+		// from normalizedToolCalls so they never reach the assistant
+		// message, the tool-result list, or the session store.
+		// A single compact rejection message is injected instead.
+		var interviewRejected []string
+		if isPlanPreExecution(agent.ContextBuilder.GetPlanStatus()) {
+			allowed := normalizedToolCalls[:0] // reuse backing array
+			for _, tc := range normalizedToolCalls {
+				if isToolAllowedDuringInterview(tc.Name, tc.Arguments) {
+					allowed = append(allowed, tc)
+				} else {
+					interviewRejected = append(interviewRejected, tc.Name)
+				}
+			}
+			normalizedToolCalls = allowed
+			if len(interviewRejected) > 0 {
+				logger.InfoCF("agent", "Interview mode: rejected tool calls",
+					map[string]any{
+						"agent_id": agent.ID,
+						"rejected": interviewRejected,
+					})
+				messages = append(messages, providers.Message{
+					Role:    "user",
+					Content: interviewRejectMessage,
+				})
+			}
+			// If all tool calls were rejected, skip to next iteration.
+			if len(normalizedToolCalls) == 0 {
+				continue
+			}
+		}
+
 		// Log tool calls
 		toolNames := make([]string, 0, len(normalizedToolCalls))
 		for _, tc := range normalizedToolCalls {
@@ -1734,15 +1774,8 @@ func (al *AgentLoop) runLLMIteration(
 				}
 			}
 
-			// Block non-allowed tools during plan interview mode.
-			// Only read-type tools and MEMORY.md writes are permitted.
 			toolStart := time.Now()
-			var toolResult *tools.ToolResult
+			toolResult := agent.Tools.ExecuteWithContext(ctx, tc.Name, tc.Arguments, opts.Channel, opts.ChatID, asyncCallback)
-			if isPlanPreExecution(agent.ContextBuilder.GetPlanStatus()) && !isToolAllowedDuringInterview(tc.Name, tc.Arguments) {
-				toolResult = tools.ErrorResult("Interview mode: only read tools and MEMORY.md edits are allowed. Focus on asking questions and updating the plan.")
-			} else {
-				toolResult = agent.Tools.ExecuteWithContext(ctx, tc.Name, tc.Arguments, opts.Channel, opts.ChatID, asyncCallback)
-			}
 			toolDuration := time.Since(toolStart)
 
 			// Update tool log entry with result
@@ -2016,20 +2049,21 @@ func (al *AgentLoop) ListSkills() []skills.SkillInfo {
 }
 
 // GetPlanInfo returns plan state from the default agent's memory store.
-func (al *AgentLoop) GetPlanInfo() (hasPlan bool, status string, currentPhase, totalPhases int, display string) {
+func (al *AgentLoop) GetPlanInfo() (hasPlan bool, status string, currentPhase, totalPhases int, display string, memory string) {
 	agent := al.registry.GetDefaultAgent()
 	if agent == nil {
-		return false, "", 0, 0, "No agent available."
+		return false, "", 0, 0, "No agent available.", ""
 	}
 	mem := agent.ContextBuilder.Memory()
 	if mem == nil {
-		return false, "", 0, 0, "No memory store."
+		return false, "", 0, 0, "No memory store.", ""
 	}
 	hasPlan = mem.HasActivePlan()
 	status = mem.GetPlanStatus()
 	currentPhase = mem.GetCurrentPhase()
 	totalPhases = mem.GetTotalPhases()
 	display = mem.FormatPlanDisplay()
+	memory = mem.ReadLongTerm()
 	return
 }
 

pkg/miniapp/miniapp.go

@@ -45,6 +45,7 @@ type PlanInfo struct {
 	TotalPhases  int         `json:"total_phases"`
 	Display      string      `json:"display"`
 	Phases       []PlanPhase `json:"phases"`
+	Memory       string      `json:"memory"`
 }
 
 // SessionInfo represents an active session entry for the API response.

pkg/miniapp/static/index.html

@@ -403,6 +403,96 @@
     border-color: var(--btn);
   }
 
+  /* MEMORY.md raw display */
+  .memory-view {
+    font-family: 'SF Mono', 'Menlo', 'Consolas', monospace;
+    font-size: 13px;
+    line-height: 1.6;
+    white-space: pre-wrap;
+    word-break: break-word;
+    padding: 14px 16px;
+    background: var(--glass-bg);
+    border: 1px solid var(--glass-border);
+    border-radius: 16px;
+    box-shadow: 0 1px 3px var(--glass-shadow);
+    -webkit-backdrop-filter: blur(12px);
+    backdrop-filter: blur(12px);
+  }
+
+  .memory-view .md-h1 {
+    font-size: 18px;
+    font-weight: 700;
+    margin: 16px 0 8px;
+    font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
+  }
+  .memory-view .md-h1:first-child { margin-top: 0; }
+
+  .memory-view .md-h2 {
+    font-size: 15px;
+    font-weight: 700;
+    margin: 14px 0 6px;
+    color: var(--link);
+    font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
+  }
+
+  .memory-view .md-h3 {
+    font-size: 14px;
+    font-weight: 600;
+    margin: 12px 0 4px;
+    font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
+  }
+
+  .memory-view .md-checkbox {
+    margin: 2px 0;
+  }
+
+  .memory-view .md-checkbox-icon {
+    display: inline-block;
+    width: 16px;
+    height: 16px;
+    border-radius: 4px;
+    border: 1.5px solid var(--hint);
+    vertical-align: middle;
+    margin-right: 6px;
+    position: relative;
+    top: -1px;
+  }
+
+  .memory-view .md-checkbox-icon.checked {
+    background: var(--done);
+    border-color: var(--done);
+  }
+
+  .memory-view .md-checkbox-icon.checked::after {
+    content: '';
+    position: absolute;
+    left: 4px;
+    top: 1px;
+    width: 4px;
+    height: 8px;
+    border: solid #fff;
+    border-width: 0 1.5px 1.5px 0;
+    transform: rotate(45deg);
+  }
+
+  .memory-view .md-quote {
+    border-left: 3px solid var(--hint);
+    padding-left: 10px;
+    color: var(--hint);
+    margin: 4px 0;
+  }
+
+  .memory-view .md-bullet {
+    margin: 2px 0;
+    padding-left: 12px;
+    text-indent: -12px;
+  }
+
+  .memory-view .md-bullet::before {
+    content: '\2022 ';
+    color: var(--hint);
+  }
+
 </style>
 </head>
 <body>
@@ -584,13 +674,55 @@ function renderPlanFromData(data) {
     '<div style="color:var(--hint);margin-top:4px">Phase ' + data.current_phase + ' / ' + data.total_phases + '</div>' +
   '</div>';
 
-  if (data.phases && data.phases.length > 0) {
+  if (data.status === 'interviewing' || data.status === 'review') {
-    html += renderPhases(data.phases, data.current_phase);
+    // Show raw MEMORY.md content during interview/review
+    if (data.memory) {
+      html += '<div class="memory-view">' + renderSimpleMarkdown(data.memory) + '</div>';
+    }
+  } else {
+    // Show structured phase/step list during executing
+    if (data.phases && data.phases.length > 0) {
+      html += renderPhases(data.phases, data.current_phase);
+    }
   }
 
   el.innerHTML = html;
 }
 
+function renderSimpleMarkdown(text) {
+  var lines = text.split('\n');
+  var out = [];
+  for (var i = 0; i < lines.length; i++) {
+    var line = lines[i];
+    // Headings
+    if (/^### /.test(line)) {
+      out.push('<div class="md-h3">' + escapeHtml(line.slice(4)) + '</div>');
+    } else if (/^## /.test(line)) {
+      out.push('<div class="md-h2">' + escapeHtml(line.slice(3)) + '</div>');
+    } else if (/^# /.test(line)) {
+      out.push('<div class="md-h1">' + escapeHtml(line.slice(2)) + '</div>');
+    // Checkboxes
+    } else if (/^- \[x\] /.test(line)) {
+      out.push('<div class="md-checkbox"><span class="md-checkbox-icon checked"></span>' + escapeHtml(line.slice(6)) + '</div>');
+    } else if (/^- \[ \] /.test(line)) {
+      out.push('<div class="md-checkbox"><span class="md-checkbox-icon"></span>' + escapeHtml(line.slice(6)) + '</div>');
+    // Blockquote
+    } else if (/^> /.test(line)) {
+      out.push('<div class="md-quote">' + escapeHtml(line.slice(2)) + '</div>');
+    // Bullet list
+    } else if (/^- /.test(line)) {
+      out.push('<div class="md-bullet">' + escapeHtml(line.slice(2)) + '</div>');
+    // Empty line
+    } else if (line.trim() === '') {
+      out.push('<br>');
+    // Plain text
+    } else {
+      out.push('<div>' + escapeHtml(line) + '</div>');
+    }
+  }
+  return out.join('');
+}
+
 async function loadPlan() {
   var loading = document.getElementById('plan-loading');
   var el = document.getElementById('plan-content');