home / skills / abdullahbeam / nexus-design-abdullah / beam-debug-issue-tasks

beam-debug-issue-tasks skill

/00-system/skills/beam/beam-debug-issue-tasks

This skill helps you diagnose and debug failed Beam.ai tasks by tracing Langfuse outputs and generating actionable reports.

npx playbooks add skill abdullahbeam/nexus-design-abdullah --skill beam-debug-issue-tasks

Review the files below or copy the command above to add this skill to your agents.

Files (2)
SKILL.md
4.0 KB
---
name: beam-debug-issue-tasks
description: Debug failed/issue tasks from Beam.ai using Langfuse traces. Load when user says "debug issue tasks", "check failed tasks", "why did task fail", "task errors", "debug agent", or needs to investigate task failures.
---

# Beam Debug Issue Tasks

**Debug failed Beam.ai tasks using Langfuse traces.**

## When to Use

- Diagnose why a task failed, stopped, or needs input
- Find root cause from Langfuse trace reasoning
- Generate debug reports for documentation or handoff

---

## Prerequisites

`.env` file at project root:

```
# Beam.ai - BID instance
BEAM_API_KEY=your_bid_api_key
BEAM_WORKSPACE_ID=your_bid_workspace_id

# Beam.ai - Prod instance
BEAM_API_KEY_PROD=your_prod_api_key
BEAM_WORKSPACE_ID_PROD=your_prod_workspace_id

# Langfuse (self-hosted)
LANGFUSE_PUBLIC_KEY=pk-lf-...
LANGFUSE_SECRET_KEY=sk-lf-...
LANGFUSE_HOST=https://tracing.beamstudio.ai
```

**Dependencies**: `pip install requests python-dotenv`

---

## Quick Start

```bash
# List issue tasks (default: last 1 day, BID workspace)
python 03-skills/beam-debug-issue-tasks/scripts/debug_issue_tasks.py <agent_id>

# Debug specific task with full trace analysis
python 03-skills/beam-debug-issue-tasks/scripts/debug_issue_tasks.py <agent_id> --task-id <task_id>

# Use prod workspace
python 03-skills/beam-debug-issue-tasks/scripts/debug_issue_tasks.py <agent_id> --workspace prod
```

---

## Workspaces

| Workspace | API Endpoint | Langfuse Project |
|-----------|--------------|------------------|
| `bid` (default) | api.bid.beamstudio.ai | cmauxbgww000582ry4644c2qr |
| `prod` | api.beamstudio.ai | clw5gbhuy0003u3rv4jzzoesh |

---

## Issue Statuses

Tasks are flagged as "issue" if status is:
- `FAILED` - Execution failed
- `ERROR` - Processing error
- `STOPPED` - Condition failed
- `CANCELLED` - User cancelled
- `TIMEOUT` - Execution timeout
- `USER_INPUT_REQUIRED` - Missing input data

---

## Debug Reports

Reports saved to: `04-workspace/agents/{agent_name}/debug/`

**Format**: Smart Brevity (headline, takeaway, why it matters, details, fix)

**Key spans analyzed**:
- `ParameterSelection/v2` - How parameters were matched
- `ExecuteGPT_Tool/v1` - Tool execution reasoning
- `NodeSelection:EdgeEvaluation/v1` - Routing decisions
- `TaskSuccessCriteriaCheck/v1` - Why task stopped

---

## CLI Reference

| Flag | Description | Default |
|------|-------------|---------|
| `agent_id` | Beam agent ID (required) | - |
| `--workspace`, `-w` | Workspace: bid or prod | bid |
| `--days`, `-d` | Look back period (1, 3, 7, 14, 30) | 1 |
| `--task-id`, `-t` | Debug specific task ID | - |
| `--summary`, `-s` | Show grouped summary | false |
| `--limit`, `-l` | Max tasks to show | 10 |
| `--output`, `-o` | Save to JSON file | - |
| `--no-trace` | Skip Langfuse lookup | false |

---

## Example Output

### Debug Report (Smart Brevity)

```markdown
# Task stopped: condition failed

Checklist evaluation: subfolder must equal 'Schreiben Schuldner' but was null.

**Why it matters**: This task did not complete successfully and may need attention.

**The details**:
- **Status**: `STOPPED`
- **Task**: `ab3cbbb8-28da-41aa-b726-25931d14d7d4`
- **Latency**: 159.6s
- **Cost**: $0.1043

**Key spans**:
- NodeSelection:EdgeEvaluation/v1 (23.4s)
- TaskSuccessCriteriaCheck/v1 (7.2s)

**Root cause**:
> The criterion is not met because subfolder is not set to required value.

**Fix**: Review the condition that stopped execution. Check if input data meets requirements.
```

---

## Langfuse Links

Each report includes direct links:
- **Session URL**: All traces for the task
- **Trace URL**: Specific execution with full details

---

## Error Handling

| Error | Solution |
|-------|----------|
| `BEAM_API_KEY not found` | Add to .env |
| `Invalid workspace` | Check workspace parameter (bid/prod) |
| `No traces found` | Verify agent has Langfuse integration |
| `401 Unauthorized` | Verify API keys |

---

## Related Skills

- `beam-get-agent-analytics` - Performance metrics
- `beam-create-agent-task` - Create test tasks
- `beam-list-agents` - List available agents

Overview

This skill debugs failed or problematic Beam.ai tasks by analyzing Langfuse traces and generating concise debug reports. It helps pinpoint root causes, surfacing critical spans, status, latency, and actionable fixes for each issue. Load it when you need to investigate task failures, timeouts, or missing inputs.

How this skill works

The skill fetches issue tasks from a specified Beam workspace, optionally limited by days or a specific task ID. It retrieves Langfuse traces for each task, analyzes key spans (parameter selection, tool execution, routing, success checks), and produces a Smart Brevity debug report with root cause and remediation steps. Reports are saved per agent for review and sharing.

When to use it

  • Investigating why a Beam task failed, stopped, or timed out
  • Finding the root cause using Langfuse trace reasoning
  • Producing a concise debug report for handoff or documentation
  • Verifying whether missing inputs or condition checks blocked progress
  • Checking tool execution failures or routing decisions

Best practices

  • Provide agent_id and use --task-id to focus analysis on a single problematic run
  • Start with a short lookback (1–3 days) then widen if needed
  • Include Langfuse keys and correct workspace in .env before running
  • Use the --summary flag to scan multiple issues quickly, then drill into full traces
  • Save reports to the agent debug folder for consistent handoffs

Example use cases

  • List issue tasks for the last day to identify recent failures in a development agent
  • Run a full trace analysis on a specific task to extract the exact span and error message
  • Generate a Smart Brevity report to share with engineering or product teams
  • Compare issue patterns across bid and prod workspaces to spot environment-specific bugs
  • Quickly confirm whether a task stopped due to missing input or an execution error

FAQ

What environment variables are required?

You need Beam API keys and workspace IDs for bid and/or prod, plus Langfuse public/secret keys and host in a .env file.

How do I limit the scope to production workspace?

Pass --workspace prod to target the production API endpoint and Langfuse project.

What does the report include?

Each report includes status, latency, cost, key spans analyzed, root cause summary, and suggested fixes, plus Langfuse trace links.