hermes - 💡(How to fix) Fix Feature: Frustration detection and /progress-report command — non-interrupting task status checks [1 participants]

CATEGORY: presence_probe
  - "are you there", "you there", "you alive", "still there"
  - "anyone", "hello", "hey", "yo", "hi"
  - "ping", "beep", "boop"
  - Repeated single messages in quick succession (2+ identical messages within 60s)

CATEGORY: status_request
  - "what's going on", "whats going on", "what's happening"
  - "status", "progress", "report"
  - "how's it going", "how is it going"
  - "any update", "updates", "progress report"
  - "report back", "give me a status"
  - "where are we", "where are we at"

CATEGORY: impatience_indicator
  - "hello??", "hey???", "???", "..."
  - "hello" followed by another "hello" within 120s
  - "are you there" sent after a long silence (>2 min with no output)
  - ALL CAPS messages: "HEY", "STATUS", "REPORT"

CATEGORY: concern_probe
  - "did it crash", "is it stuck", "did it fail"
  - "is this done yet", "how much longer"
  - "why is this taking so long"
  - "timeout", "timed out", "error"

CATEGORY: gentle_nudge
  - "sup", "what's up", "whatsup", "whatup"
  - "how's it looking", "anything new"
  - "finished?", "done yet"

---

def classify_message(text, context):
    """
    Returns: 'frustration_probe' | 'steer_request' | 'normal_message'
    """
    text_lower = text.strip().lower()
    
    # Quick rejection: if the message is long (>50 chars) or contains
    # specific action verbs, it's probably a real steer request
    if len(text_lower) > 50 and not any(p in text_lower for p in SHORT_PROBES):
        return 'normal_message'
    
    # If there's an active long-running task, these are progress probes
    if active_task_exists():
        for pattern_set in FRUSTRATION_PATTERNS:
            if any(p in text_lower for p in pattern_set):
                return 'frustration_probe'
    
    # Without an active task, short messages are just greetings
    return 'normal_message'

---

"hey, stop the build"        → steer_request (has an action verb + target)
"are you there? cancel it"   → steer_request (frustration opener but real command)
"what's going on with X?"    → steer_request (asking about a topic, not progress)
"status of the deploy"       → steer_request (asking about a specific task, not current task)

---

/progress-report
/progressreport
/whatsgoingon
/whats-up
/whatsup
/status
/report

---

def get_progress_report():
    report = {}
    
    # 1. Active process check
    if terminal_process_running():
        proc = get_active_process()
        report["status"] = "running"
        report["command"] = proc.command  # truncated
        report["started_at"] = proc.started_at
        report["elapsed"] = now() - proc.started_at
        report["last_output"] = proc.last_output_line
        report["last_output_at"] = proc.last_output_timestamp
        report["cpu_usage"] = proc.cpu_percent
        report["memory_usage"] = proc.memory_mb
        
        # Detect stall: no output for >2 minutes on a process that was producing output
        if proc.was_producing_output and (now() - proc.last_output_timestamp > 120):
            report["stall_detected"] = True
            report["stall_duration"] = now() - proc.last_output_timestamp
            report["stall_context"] = proc.last_output_lines(20)
    
    # 2. Cron job check
    active_crons = get_running_cron_jobs()
    if active_crons:
        for job in active_crons:
            report["cron_jobs"].append({
                "name": job.name,
                "job_id": job.job_id,
                "started_at": job.started_at,
                "elapsed": now() - job.started_at,
                "model": job.model,
                "tools_used": job.tools_accessed,
                "last_status": job.last_status_update
            })
    
    # 3. Delegated agent check
    if delegated_agent_active():
        agent = get_active_delegated_agent()
        report["delegated_agent"] = {
            "goal": agent.goal_summary,
            "started_at": agent.started_at,
            "elapsed": now() - agent.started_at,
            "current_step": agent.current_step,
            "tools_called": agent.tools_call_count
        }
    
    # 4. Queue check
    pending_messages = get_pending_messages()
    if pending_messages:
        report["queued_messages"] = len(pending_messages)
        report["oldest_pending"] = pending_messages[0].received_at
    
    return report

---

⏳ **Still working on it**

🔍 **GitHub Issue Monitor** (cron job)
⏱ Running for 4m 23s
🔧 Tools called: 12 (terminal, web_search, gh api)
📊 Last activity: 8s ago — fetching comments on hermes-agent#18289

Estimate: based on tool call rate, ~3-5 minutes remaining.

---

⚠️ **Task may be stuck**

🔍 **GitHub Issue Monitor** (cron job)
⏱ Running for 12m 05s
🔧 Tools called: 8 (last one 8m ago)
📊 Last output: "Checking comments on OmniRoute#1814..."

📋 **Possible causes:**
  • API rate limit hit (GitHub 5000/hr, 8 calls so far — unlikely)
  • Model timeout waiting for response
  • Process hung on network call

🔧 **Auto-action:** Spawning diagnostic check...
[diagnostic results follow in next message]

---

✅ **Task finished** — response wasn't delivered

The GitHub Issue Monitor completed 3m ago but the result wasn't sent.
Re-delivering now:

📊 **GitHub Issue Activity** (last 8h)
[...full result...]

---

😴 **Nothing running**

No active tasks, cron jobs, or agents. Last activity:
  • GitHub Issue Monitor — completed 2h ago ✅
  • Software Cache Refresh — scheduled in 5 days

Queued messages: 0

All clear. Need something?

---

ROLE: Stall Diagnostician
MODEL: aux (flash-tier)
TIMEOUT: 30 seconds
TOOLSETS: [terminal]

TASK:
1. Check if the process is still alive: ps aux | grep {pid}
2. Check last 50 lines of output: process log
3. Check for common failure modes:
   a. API rate limit: grep for "rate limit", "429", "quota"
   b. Network timeout: grep for "timeout", "ETIMEDOUT", "ECONNREFUSED"
   c. Model error: grep for "error", "failed", "exception", "Traceback"
   d. Authentication: grep for "unauthorized", "401", "403"
   e. Disk space: df -h (running out of space causes silent hangs)
   f. Memory pressure: free -h (OOM kills don't always produce output)
   g. DNS issues: check if the machine can resolve hosts
4. Check system resources: top -bn1 | head -20
5. Return structured diagnosis

OUTPUT:
{
  "stall_cause": "api_rate_limit",
  "confidence": 0.85,
  "evidence": "GitHub API returned 403 with 'rate limit exceeded' at 17:23:41",
  "suggestion": "Wait for rate limit window to reset (~15 min) or reduce API calls",
  "auto_recovery_possible": true,
  "auto_recovery_action": "wait_and_retry"
}

---

⚠️ **Heads up** — GitHub Issue Monitor seems stuck

Been running for 15m with no activity in the last 8 minutes.
Likely cause: GitHub API rate limit hit.

I'll retry automatically when the window resets (~22:45).

React 👍 to acknowledge, or 🛑 to cancel the job.

---

User message arrives
        │
        ▼
┌─────────────────────┐
│  Pattern Classifier │
│  (regex, <50 chars,  │
│   active task check) │
└─────────┬───────────┘
          │
          ├── "frustration_probe" ──► Progress Reporter (non-interrupting)
          │                                    │
          │                                    ├── Report from internal state
          │                                    ├── Spawn diagnostic if stalled
          │                                    └── Send response immediately
          │
          ├── "/progress-report" ──► Same Progress Reporter
          │
          └── "normal_message" ──► Standard agent queue

---

agent:
  frustration_detection:
    enabled: true  # on by default — this should just work
    
    # Pattern sensitivity
    detect_short_probes: true     # "hey", "ping", "???"
    detect_repeated_messages: true # same message 2+ times in 60s
    detect_all_caps: true         # "STATUS", "REPORT"
    
    # Don't classify as frustration if message contains these
    action_verbs: ["stop", "cancel", "abort", "kill", "skip", "change", "switch", "do", "run", "make", "create", "fix", "update"]
    
    # Response style
    response_includes_estimate: true   # rough time remaining
    response_includes_tools_used: true # what tools have been called
    response_includes_next_step: true  # what the agent is about to do
    
  progress_report:
    command_aliases:
      - /progress-report
      - /progressreport
      - /whatsgoingon
      - /whats-up
      - /whatsup
      - /status
      - /report
    
  stall_detection:
    enabled: true
    check_interval_seconds: 30
    
    # Stall thresholds
    no_output_threshold_seconds: 120
    no_tool_calls_threshold_seconds: 300
    no_status_update_threshold_seconds: 600
    cpu_zero_threshold_seconds: 60
    repeated_output_threshold: 3
    
    # Auto-diagnosis
    auto_diagnose: true
    diagnostic_model: "aux"  # use cheapest available
    
    # Auto-recovery
    auto_recovery: true
    auto_recovery_actions:
      rate_limit: "wait_and_retry"
      network_timeout: "retry_with_backoff"
      model_timeout: "fallback_to_aux"
    
    # Proactive notification
    notify_on_stall: true
    stall_notify_after_seconds: 300  # 5 min of no activity
    
    # Emoji acknowledgment of proactive notifications
    ack_emoji: "👍"   # user reacts 👍 to acknowledge
    cancel_emoji: "🛑" # user reacts 🛑 to cancel stalled job