bee-tasks

name: bee-tasks description: Stateless T4 administrative worker swarm. Spawn ephemeral administrative worker agents (T4) to execute predefined task templates with no memory persistence. Use when delegating cleanup, organization, validation, or administrative work that requires no context retention or workspace persistence. Supports task template execution, queue management, and handoff from T1-T3 agents. NOT for complex reasoning, research, or tasks requiring memory/state.

Bee-Tasks: T4 Administrative Worker Swarm

Ephemeral worker swarm for administrative tasks. These agents are stateless, disposable, and designed for high-throughput task execution.

Core Design Principles

When to Use

USE when:

• Administrative cleanup tasks
• File organization operations
• Batch validation checks
• Data formatting/transformation
• Status aggregation across systems
• Routine maintenance scripts

DON'T use when:

• Multi-step reasoning required
• Context must persist across operations
• Workspace needed for intermediate files
• Research or analysis tasks
• Creative or open-ended work

Tier Model Context

Task Templates

Template Format

task_template:
  name: "task_name"
  description: "What this task does"
  inputs:
    - param1: "description"
    - param2: "description"
  outputs:
    - result: "expected output format"
  timeout: "60s"

Standard Templates

cleanup-empty-files

Remove empty files from a directory tree.

inputs:
  - path: "root directory to scan"
outputs:
  - deleted: "list of removed files"
  - skipped: "count of non-empty files"

validate-json-batch

Validate all JSON files in a directory.

inputs:
  - path: "directory containing .json files"
outputs:
  - valid: "count of valid files"
  - invalid: "list of files with errors"

aggregate-logs

Collect and summarize log entries across files.

inputs:
  - pattern: "glob pattern for log files"
  - level: "log level to filter (ERROR, WARN, etc.)"
outputs:
  - summary: "count per level"
  - samples: "representative entries"

organize-by-date

Move files into date-based subdirectories.

inputs:
  - source: "source directory"
  - date_format: "YYYY-MM or YYYY-MM-DD"
outputs:
  - moved: "count of files organized"
  - structure: "resulting directory layout"

Queue Management

Task Submission

{
  "queue": "admin",
  "task": "cleanup-empty-files",
  "params": {
    "path": "/tmp/workspace/cache"
  },
  "priority": "normal",
  "deadline": "2026-03-24T12:00:00Z"
}

Queue Priorities

Worker Scaling

• Idle: 0 workers (cold start acceptable for T4)
• Queue depth <10: 1 worker
• Queue depth 10-50: 2 workers
• Queue depth >50: Scale to 4 workers

Handoff from T1-T3

Initiating Handoff

T1-T3 agents can spawn T4 workers by:

1. Define the task:

   handoff:
     from: "T2-agent-id"
     task_template: "aggregate-logs"
     params:
       pattern: "*.log"
       level: "ERROR"
     callback: "T2-agent-id"
   

2. Return context: T4 receives only the params; no session history

3. Receive result: Output delivered to callback handler

Example Handoff Pattern

T2 (research agent):
  ↓ Identifies need: "Check all log files for errors"
  ↓ Spawns T4 worker with params
  
T4 (bee worker):
  ↓ Executes: aggregates ERROR entries
  ↓ Returns: summary + samples
  
T2 (research agent):
  ↓ Receives result
  ↓ Incorporates into analysis

Execution Protocol

Spawn Command

# Spawn T4 worker for administrative task
sessions_spawn(
  model="custom-api-staging-ytlailabs-tech/ilmu-mini-free-v2",
  task_template="cleanup-empty-files",
  params={"path": "/var/tmp"},
  timeout=60,
  memory=false,
  workspace=false
)

Return Format

T4 workers MUST return results in this structure:

{
  "status": "success|failure|timeout",
  "task": "template_name",
  "duration_ms": 1234,
  "result": {
    // template-specific output
  },
  "errors": []
}

Monitoring

Metrics to Track

• Task completion rate (target: >99%)
• Average execution time per template
• Queue depth over time
• Worker spawn time (cold start latency)

Alert Thresholds

Best Practices

1. Keep templates atomic - One clear operation per template
2. Validate inputs - Fail fast on missing/invalid params
3. Return structured data - JSON outputs preferred
4. No side effects - T4 should not modify state outside task scope
5. Idempotent operations - Same params = same result

Security

• T4 workers have limited scope
• No access to credentials store
• No cross-agent communication
• Read-only on sensitive paths unless explicitly permitted