Execution Monitoring
v10.0.178+ (floating task board).
After the plan is written, the agent follows your natural-language direction to continue, pause, or revise execution. You can monitor progress and intervene at any time; there is no blanket plan-approval state to operate first. For long-running work, DesireCore uses the floating task board when responsibility, blockers, cross-agent coordination, or durable progress visibility genuinely need tracking.
Real-Time Display of Execution Process
When a task is executing, you will see a real-time updated progress panel:
┌──────────────────────────────────────────────────────┐
│ 🔄 Executing: Review procurement contract │
├──────────────────────────────────────────────────────┤
│ │
│ ✅ Step 1/6 Parse contract file Done (0.8s) │
│ ✅ Step 2/6 Check penalty percentage Done (0.2s) │
│ └─ Result: 15%, meets standard (≤20%) │
│ ✅ Step 3/6 Check payment terms Done (0.2s) │
│ └─ Result: ⚠️ 15 days, below standard (≥30 days) │
│ 🔄 Step 4/6 Analyze imported equipment clauses In progress... │
│ └─ Analyzing clauses 12-15... │
│ ⏳ Step 5/6 Comprehensive risk assessment Waiting │
│ ⏳ Step 6/6 Wait for confirmation before sending Waiting │
│ │
│ Total progress: ━━━━━━━━━━━━░░░░░░░░ 60% │
│ Time elapsed: 1 min 15 sec | Est. remaining: 1-2 min │
│ │
│ ┌──────────┐ ┌──────────┐ │
│ │ ⏸ Pause │ │ ⏹ Stop │ │
│ └──────────┘ └──────────┘ │
│ │
└──────────────────────────────────────────────────────┘
Step Status Indicators
| Status | Icon | Meaning |
|---|---|---|
| Pending | 🔹 | Task created but not started |
| Waiting | ⏳ | Queued (waiting on a dependency or an earlier step to finish) |
| In Progress | 🔄 | Currently processing |
| Completed | ✅ | Step executed successfully |
| Needs Confirmation | 🚪 | Human Gate, waiting for your confirmation |
| Warning | ⚠️ | Execution successful but problem found |
| Failed | ❌ | Execution failed |
| Discarded | 🗑️ | Task soft-deleted (still visible in the audit log) |
Task List (Transparent Progress)
A task list is not a state machine for the Plan, and it does not mechanically duplicate every Plan step. The agent creates task items only when work needs explicit ownership, blocker tracking, cross-agent coordination, or durable visibility of key progress. Simple sequencing remains in the Plan file. You can still ask “where are we?” or “what’s left?” at any time.
What the Task List Looks Like
Each task has a clear title, status, and links to related context:
📋 Task List (agent "Contract Review Assistant")
🔄 #1 Parse contract file, extract key clauses
In progress · linked to .desirecore/plans/<plan-id>/PLAN.md
🔹 #2 Check penalty percentage
Pending · depends on #1 completing
🔹 #3 Check payment terms
Pending · depends on #1 completing
🔹 #4 Analyze imported equipment clauses (adaptive)
Pending · risk: med · depends on #1 #2 #3
🔹 #5 Comprehensive risk assessment (adaptive)
Pending · risk: med · needs your confirmation
🔹 #6 Send review report
Pending · risk: high · needs your confirmation
What task states mean
| State | Meaning | What you can do |
|---|---|---|
| Not started | Work has not begun | Check the assignee and prerequisites |
| In progress | Work is underway | Inspect the latest update and saved outputs |
| Blocked | Input, permission, or a prerequisite is missing | Read the reason and provide what is needed |
| Awaiting confirmation | Output has been submitted for acceptance | Open the output, accept it, or request changes |
| Completed | The task has completed under its rules | Check the final deliverables |
| Failed | Execution did not finish successfully | Inspect the error and saved outputs before retrying |
The agent and task board maintain these states; they do not report every background command in real time. If updates stop, ask for progress or check Task Manager to see whether execution is still running.
Task Dependencies
A task can declare "you must finish X before me". For example, "run database migration" depends on "backup complete"—if the dependency hasn't finished, the agent won't start the dependent step, preventing out-of-order execution.
Dependencies appear in the list as "depends on #X" so you can see the full task graph at a glance.
Deviations from the Plan Reflect in the List
If the agent needs to adjust steps mid-flight, it will:
- Proactively tell you "I'm going to deviate from the original plan, because X"
- Incorporate your feedback through natural conversation; wait only when you explicitly require confirmation, otherwise update the relevant task items
- Append the deviation log to the
## Execution Notessection at the end of the plan file
This way, looking at the plan file + task list later gives you a complete reconstruction of "what actually happened".
The current conversation and the Global Task Board
The conversation task list helps track current work. For work across conversations or multiple agents, open Global Task Board from the left navigation and browse by board or assignee. Board permissions determine which tasks can be viewed or changed; team members also follow assignment and review rules.
Registering a task does not start execution. See Long-running tasks for the full sequence.
Deterministic Steps vs. Adaptive Steps
Execution steps may follow defined rules or involve AI judgment. Check the step type and the basis for its result.
Deterministic Step
- Characteristics: Clear rules, certain results
- Execution Method: Executed precisely like a program
- Predictability: Execute 100 times, 100 times same result
- Examples: Check if value exceeds threshold, extract specific fields, format conversion
⚙️ [Deterministic] Check penalty percentage
Rule: Penalty ≤ 20% → Compliant
Input: Contract clause 8, penalty = 15%
Result: ✅ Compliant (15% ≤ 20%)
Adaptive Step
- Characteristics: Requires AI understanding and judgment
- Execution Method: Based on contextual reasoning
- Predictability: Different inputs may have different reasoning paths
- Examples: Analyze contract semantics, assess comprehensive risk, generate reports
🧠 [Adaptive] Analyze imported equipment clauses
Basis: Contract clauses 12-15 content
Reasoning process:
- Clause 12 mentions "imported equipment" but doesn't specify country of origin
- Clause 13 has installation acceptance terms but doesn't mention Chinese instruction manual
- According to learned rules: imported equipment must have Chinese instruction manual clause
Result: ⚠️ Missing Chinese instruction manual clause, suggest adding
Deterministic steps follow defined rules, but their inputs, rules, or environment may still be incorrect. Adaptive steps also involve AI judgment. Check results in both cases, and inspect the reasoning behind judgment-based results.
Stopping and Continuing Execution
| Your goal | Action | What to check next |
|---|---|---|
| Interrupt execution | Use the current run's stop action, or ask the agent to pause | Check Task Manager for background commands that are still running |
| Continue unfinished work | Ask the agent to inspect its plan and saved outputs before continuing | Completed steps, remaining work, and commands still running |
| Retry a failed step | Read the failure reason before requesting a retry | Check for saved files and submitted external requests to avoid duplicates |
| Restore local changes | Open Rewind and Checkpoint | Check which files are covered and what will be restored |
Stopping execution does not guarantee cancellation of requests already accepted by external services. Retry, skip, and continuation options depend on the task; some operations cannot resume from the point of interruption.
Exception Handling During Execution
The workflow panel also shows failures at run level, including startup or connection errors before any node begins. Absence of a failed node does not mean success. Read the run failure message and check completed actions before reconnecting or retrying.
When the agent encounters problems, it takes different handling approaches based on the situation:
Automatic Retry
For temporary errors like network timeouts, the agent automatically retries:
⚠️ Step 4 execution failed: File read timeout
Retrying automatically... (1/3)
✅ Retry successful
Proactive Reporting
For problems it cannot solve on its own, the agent proactively tells you:
Agent: "Step 4 encountered a problem:
⚠️ Contract file clause 13 references an attachment (Attachment C: Equipment Specification),
but I didn't find this attachment.
Please choose:
1. Provide Attachment C file
2. Skip this step, mark 'missing Attachment C' in report
3. Pause task, I'll handle it later"
Inserting New Instructions During Execution
You can supplement instructions to the agent at any time during execution:
You: "Wait, clause 12's confidentiality period also needs special attention,
if it exceeds 5 years mark it as a risk."
Agent: "Received, I will add confidentiality period check in subsequent steps.
Current step unaffected, will incorporate in step 5 (comprehensive assessment)."
Approval and Waiting in Delegated Tasks
A child Agent may request approval while executing a command or tool. DesireCore displays the request and final decision in the child conversation and mirrors both into the parent Agent conversation that started the delegation. Notifications also prefer the parent conversation, so you do not need to switch panels just to approve the child operation.
| Approval mode | Waiting behavior during delegation |
|---|---|
| AI Approval | You can decide within 30 seconds; at expiry, the AI suggestion approves or rejects, or the operation is denied when no suggestion is available |
| Always Ask / Allowlist miss / External Tools Only | Waits for a human decision and does not reject merely because an ordinary approval countdown elapsed |
| Allow All | Skips ordinary confirmation, while deny lists, file scope, and tool permissions remain active |
A synchronous delegation has a separate wait limit. If the parent receives a “wait timed out” result first, the child task has not necessarily ended: it may still be running in the background or waiting for your approval. The timeout notice includes an available task reference; the parent can continue waiting, ignore a later result, or terminate the child task when supported.
During execution, certain high-risk operations will trigger "Human Gate" confirmation. Go to Human Gate Confirmation Mechanism for details.
See Long-running tasks for instructions on saving stage progress, running commands in the background, and resuming from existing results after an interruption.