Beam Master

Beam Master

This is NOT a user-facing skill. It's a shared resource library referenced by Beam integration skills.

Purpose

Provides shared resources to eliminate duplication across:

• beam-connect - Meta-skill for Beam AI workspace operations
• beam-list-agents - List workspace agents
• beam-get-agent-graph - Fetch agent graph configuration
• beam-get-agent-analytics - Get agent performance metrics
• beam-create-agent-task - Create and execute agent tasks
• beam-update-graph-node - Update node configuration
• beam-test-graph-node - Test individual nodes
• beam-get-nodes-by-tool - Get nodes using specific tools
• beam-debug-issue-tasks - Debug failed tasks via Langfuse

Instead of loading this skill, users directly invoke the specific skill they need above.

Architecture: DRY Principle

Problem solved: Beam skills would have duplicated content (setup instructions, API docs, auth flow, error handling).

Solution: Extract shared content into beam-master/references/ and beam-master/scripts/, then reference from each skill.

Result: Single source of truth, reduced context per skill.

Shared Resources

All Beam skills reference these resources (progressive disclosure).

references/

setup-guide.md - Complete setup wizard

• Getting Beam API key
• Finding workspace ID
• Configuring .env file
• Token exchange flow

api-reference.md - Beam API patterns

• Base URL and authentication
• All 22 API endpoints documented
• Request/response examples
• Common curl examples

error-handling.md - Troubleshooting

• Common errors and solutions
• HTTP error codes
• Authentication issues
• Rate limiting

authentication.md - Token management

• API key to access token exchange
• Token refresh flow
• Header requirements

scripts/

Authentication & Configuration

check_beam_config.py - Pre-flight validation

python check_beam_config.py [--json]

Exit codes: 0=configured, 1=partial, 2=not configured

When to Use: Run this FIRST before any Beam operation. Use to validate API key and workspace ID are configured, diagnose authentication issues, or check if setup is needed.

setup_beam.py - Interactive setup wizard

python setup_beam.py

No arguments - runs interactively. Guides through API key setup, tests connection, saves to .env.

When to Use: Use when Beam integration needs initial setup, when check_beam_config.py returns exit code 2, or when user needs to reconfigure credentials.

get_access_token.py - Token exchange (POST /auth/access-token)

python get_access_token.py [--json]

When to Use: Use to exchange API key for access token. Called automatically by beam_client.py, but use directly when debugging authentication issues or testing token exchange.

refresh_token.py - Token refresh (POST /auth/refresh-token)

python refresh_token.py --refresh-token TOKEN [--json]

When to Use: Use when access token has expired and you have a refresh token. Typically handled automatically, but use directly for debugging token lifecycle issues.

get_current_user.py - User profile (GET /v2/user/me)

python get_current_user.py [--json]

When to Use: Use to verify authentication is working, get user profile info, or confirm workspace access after setup.

Agent Management

list_agents.py - List workspace agents (GET /agent)

python list_agents.py [--filter NAME] [--json]

When to Use: Use when user says "list agents", "show my agents", "what agents do I have", or when you need to find an agent ID for use with other scripts (tasks, graphs, analytics).

get_agent_graph.py - Get agent workflow graph (GET /agent-graphs/{agentId})

python get_agent_graph.py --agent-id AGENT [--graph-id GRAPH] [--json]

When to Use: Use when user wants to see agent workflow structure, understand node configuration, get node IDs for testing/updating, or analyze agent architecture.

Graph & Node Operations

test_graph_node.py - Test a specific node (POST /agent-graphs/test-node)

python test_graph_node.py --agent-id AGENT --node-id NODE --graph-id GRAPH [--input JSON] [--json]

When to Use: Use when debugging a specific node, testing node behavior with custom input, validating node configuration changes, or isolating issues in a workflow.

update_graph_node.py - Update node configuration (PATCH /agent-graphs/update-node)

python update_graph_node.py --node-id NODE [--objective TEXT] [--on-error STOP|CONTINUE] [--auto-retry BOOL] [--config JSON] [--json]

When to Use: Use when modifying node objectives/prompts, changing error handling behavior, enabling/disabling auto-retry, or updating node configuration programmatically.

get_nodes_by_tool.py - Find nodes by tool (GET /agent-graphs/agent-task-nodes/{toolFunctionName})

python get_nodes_by_tool.py --tool TOOL_NAME [--agent-id AGENT] [--rated] [--page N] [--page-size N] [--json]

When to Use: Use when finding all nodes that use a specific tool, analyzing tool usage patterns across agents, or gathering rated nodes for optimization training.

get_tool_output_schema.py - Get node output schema (GET /agent-tasks/tool-output-schema/{graphNodeId})

python get_tool_output_schema.py --node-id NODE [--json]

When to Use: Use when understanding what output format a node produces, debugging output parsing issues, or documenting node interfaces.

Task Operations

create_task.py - Create new agent task (POST /agent-tasks)

python create_task.py --agent-id AGENT --query "Task instructions" [--urls URL1,URL2] [--json]

When to Use: Use when user wants to run an agent, execute a task, start a job, or invoke an agent with specific instructions. This is the primary way to trigger agent execution.

get_task.py - Get task details (GET /agent-tasks/{taskId})

python get_task.py --task-id TASK [--json]

When to Use: Use when checking task status after creation, getting detailed results of a completed task, reviewing task execution history, or debugging why a task failed.

list_tasks.py - List tasks with filtering (GET /agent-tasks)

python list_tasks.py [--agent-id AGENT] [--status STATUS1,STATUS2] [--search TEXT] [--start-date DATE] [--end-date DATE] [--page N] [--page-size N] [--order FIELD:DIR] [--json]

When to Use: Use when user asks "show my tasks", "list failed tasks", "what tasks ran today", or when finding tasks by status, date range, or search criteria. Best for filtered views.

iterate_tasks.py - Paginated task iteration (GET /agent-tasks/iterate)

python iterate_tasks.py [--agent-id AGENT] [--cursor CURSOR] [--limit N] [--json]

When to Use: Use when iterating through large numbers of tasks efficiently, exporting task data, or building reports. Better than list_tasks for bulk operations due to cursor-based pagination.

retry_task.py - Retry failed task (POST /agent-tasks/retry)

python retry_task.py --task-id TASK [--json]

When to Use: Use when a task has failed and user wants to re-run it, after fixing underlying issues that caused failure, or when retrying transient errors.

get_task_updates.py - Stream real-time updates (GET /agent-tasks/{taskId}/updates - SSE)

python get_task_updates.py --task-id TASK [--timeout SECONDS] [--json]

When to Use: Use when monitoring a running task in real-time, watching for HITL (human-in-the-loop) requests, or streaming task progress updates to the user.

Task Feedback & HITL

approve_task.py - Approve HITL task (POST /agent-tasks/execution/{taskId}/user-consent)

python approve_task.py --task-id TASK [--json]

When to Use: Use when a task is waiting for human approval (HITL), agent requests user consent to proceed, or task status shows "WAITING_FOR_CONSENT".

reject_task.py - Reject task execution (POST /agent-tasks/execution/{taskId}/rejection)

python reject_task.py --task-id TASK [--reason TEXT] [--json]

When to Use: Use when user wants to cancel/reject a running task, stop task execution, or provide negative feedback that halts the task.

provide_user_input.py - Provide HITL input (PATCH /agent-tasks/execution/{taskId}/user-input)

python provide_user_input.py --task-id TASK --input "User response" [--json]

When to Use: Use when a task is waiting for human input (HITL), agent asks a question and needs user's answer, or task status shows "WAITING_FOR_USER_INPUT".

rate_task_output.py - Rate task output (PATCH /agent-tasks/execution/{taskId}/output-rating)

python rate_task_output.py --task-id TASK --node-id NODE --rating RATING [--feedback TEXT] [--expected-output TEXT] [--json]

When to Use: Use when providing feedback on task/node output quality, training the system with positive/negative examples, or improving agent performance through ratings.

Analytics & Optimization

get_analytics.py - Agent performance analytics (GET /agent-tasks/analytics)

python get_analytics.py --agent-id AGENT [--start-date YYYY-MM-DD] [--end-date YYYY-MM-DD] [--json]

When to Use: Use when user asks "how is my agent performing", "show analytics", "success rate", or when analyzing agent performance metrics over a time period.

get_latest_executions.py - Recent executions (GET /agent-tasks/latest-executions)

python get_latest_executions.py [--agent-id AGENT] [--limit N] [--json]

When to Use: Use when user asks "what ran recently", "show recent tasks", "latest executions", or when quickly checking what tasks completed recently without complex filtering.

optimize_tool.py - Start tool optimization (POST /tool/optimize/{toolFunctionName})

python optimize_tool.py --tool TOOL_NAME [--json]

When to Use: Use when user wants to improve a tool's performance, has collected rated node examples, or wants to trigger AI-driven optimization of tool prompts/behavior.

get_optimization_status.py - Check optimization status (POST /tool/optimization-status/thread/{threadId})

python get_optimization_status.py --thread-id THREAD [--json]

When to Use: Use after starting an optimization with optimize_tool.py to check progress, see if optimization is complete, or get optimization results.

File Operations

download_context_file.py - Download agent context file (GET /agent/{agentId}/context/file/{fileId}/download)

python download_context_file.py --agent-id AGENT --file-id FILE [--output PATH] [--json]

When to Use: Use when user needs to download a file attached to an agent's context, retrieve agent documentation/assets, or access files stored in agent configuration.

Intelligent Error Detection Flow

When a Beam skill fails due to missing configuration, the AI should:

Step 1: Run Config Check with JSON Output

python 00-system/skills/beam/beam-master/scripts/check_beam_config.py --json

Step 2: Parse the ai_action Field

The JSON output includes an ai_action field that tells the AI what to do:

Step 3: Help User Fix Issues

If ai_action is prompt_for_api_key:

1. Tell user: "Beam integration needs setup. I need your API key."
2. Show them: "Get one from: Beam workspace → Settings → API Keys"
3. Ask: "Paste your Beam API key here (starts with 'bm_key_'):"
4. Once they provide it, write directly to .env:

   # Edit .env file to add:
   BEAM_API_KEY=bm_key_their_key_here
   BEAM_WORKSPACE_ID=their-workspace-id
   

5. Re-run config check to verify

JSON Output Structure

{
  "status": "not_configured",
  "exit_code": 2,
  "ai_action": "prompt_for_api_key",
  "missing": [
    {"item": "BEAM_API_KEY", "required": true, "location": ".env"}
  ],
  "fix_instructions": [...],
  "env_template": "BEAM_API_KEY=bm_key_YOUR_API_KEY_HERE\nBEAM_WORKSPACE_ID=your-workspace-id",
  "setup_wizard": "python 00-system/skills/beam/beam-master/scripts/setup_beam.py"
}

How Skills Reference This

Each skill loads shared resources only when needed (progressive disclosure):

beam-connect uses:

• check_beam_config.py (validate before any operation)
• All API scripts based on user request
• All references as needed

beam-list-agents uses:

• check_beam_config.py (validate before query)
• list_agents.py (core functionality)
• error-handling.md (troubleshooting)

beam-create-agent-task uses:

• check_beam_config.py (validate before task creation)
• create_task.py (core functionality)
• get_task_updates.py (monitor execution)
• api-reference.md (request format)

Usage Example

User says: "list my beam agents"

What happens:

1. AI loads beam-connect skill (NOT beam-master)
2. beam-connect SKILL.md says: "Run check_beam_config.py first"
3. AI executes: python beam-master/scripts/check_beam_config.py --json
4. AI executes: python beam-master/scripts/list_agents.py
5. If errors occur, AI loads: beam-master/references/error-handling.md

beam-master is NEVER loaded directly - it's just a resource library.

Environment Variables

Required in .env:

BEAM_API_KEY=bm_key_xxxxxxxxxxxxx
BEAM_WORKSPACE_ID=your-workspace-id

API Base URL

All API requests go to: https://api.beamstudio.ai

Version: 1.2 Created: 2025-12-11 Updated: 2025-12-11 Status: Production Ready

Changelog:

• v1.2: Added "When to Use" sections to all 27 scripts for AI routing guidance
• v1.1: Added comprehensive script argument documentation with usage examples and argument tables