egenskriven-workflows

Overview

EgenSkriven supports configurable workflow modes that control how strictly agents should use task tracking, and agent modes that define the level of autonomy.

Workflow Modes

Configured in .egenskriven/config.json under workflow_mode.

Strict Mode

Full enforcement of task tracking. Best for complex projects, team coordination, and audit trails.

Requirements:

• •Before starting work: Create or claim a task
• •During work: Update task status, add notes
• •After completion: Mark done, create follow-up tasks

# Check current context before starting
egenskriven context --json

# Claim a task before working
egenskriven move <task-ref> in_progress

# Complete when done
egenskriven move <task-ref> done

Light Mode (Default)

Basic tracking without ceremony. Best for solo development and rapid iteration.

Requirements:

• •Create tasks for significant work
• •Complete tasks when done
• •No structured sections required

# Create task for significant work
egenskriven add "Implement feature X" --type feature

# Complete when done
egenskriven move <task-ref> done

Minimal Mode

Agent decides when to use task tracking. Best for exploratory work and quick fixes.

Behavior:

• •No enforcement or requirements
• •EgenSkriven available but optional
• •Agent uses judgment on when tracking adds value

Agent Modes

Controls the level of agent autonomy in task operations.

Autonomous Mode

Agent executes actions directly; human reviews asynchronously.

Behavior:

• •Creates, updates, completes tasks without asking
• •Human reviews via activity history
• •Ideal for trusted agents with clear scope

Example session:

# Agent automatically creates task
egenskriven add "Fix login bug" --type bug --priority high

# Agent works and updates status
egenskriven move FIX-1 in_progress
egenskriven update FIX-1 --note "Root cause: session timeout"

# Agent completes when done
egenskriven move FIX-1 done

Collaborative Mode

Agent proposes major changes, executes minor ones.

Behavior:

• •Can read tasks and make minor updates
• •Major changes (complete, delete) require explanation
• •Agent states intent, human confirms if needed
• •Balance of autonomy and oversight

Major actions requiring explanation:

• •Completing tasks
• •Deleting tasks
• •Changing priority to urgent
• •Reassigning tasks

Supervised Mode

Agent can only read task data; outputs commands for human to execute.

Behavior:

• •Read-only access to task data
• •Outputs CLI commands as suggestions
• •Maximum control, minimum agent autonomy
• •Good for sensitive projects or new agents

Example output:

I recommend completing this task. Run:
  egenskriven move FIX-1 done

Prime Command

The prime command outputs full agent instructions for hook-based injection.

# Basic usage - outputs full instructions
egenskriven prime

# Override workflow mode
egenskriven prime --workflow strict

# Identify the agent
egenskriven prime --agent claude

Hook Integration

For agents supporting hooks (Claude Code), prime can be auto-injected:

{
  "hooks": {
    "SessionStart": [
      {
        "hooks": [
          { "type": "command", "command": "egenskriven prime" }
        ]
      }
    ]
  }
}

Session Patterns

Starting a Session

1. •Run egenskriven context --json to get project state
2. •Run egenskriven suggest --json to get recommended next task
3. •Review ready tasks with egenskriven list --ready

During Work

1. •Move task to in_progress when starting
2. •Add notes for significant progress or blockers
3. •Update status as work progresses

Completing Work

1. •Verify all acceptance criteria are met
2. •Move task to done or review as appropriate
3. •Create follow-up tasks if needed

When to Create Tasks

Create tasks for:

• •Multi-step work spanning multiple exchanges
• •Work that might be interrupted
• •Features with dependencies
• •Bugs that need tracking

Skip task creation for:

• •Simple questions or explanations
• •One-off commands
• •Trivial changes (typos, formatting)
• •Information gathering

Configuration

Check current configuration:

egenskriven config show

View workflow mode in context:

egenskriven context --json | jq '.workflow_mode'

Resume Modes

Each board can be configured with a resume mode that controls how blocked tasks are resumed after human input. This is part of the Human-AI Collaborative Workflow.

Available Modes

Manual Mode (manual)

The resume command is printed for the user to copy and run manually.

$ egenskriven resume WRK-42
Resume command for WRK-42:

  opencode run '## Task Context...' --session abc-123

Working directory: /home/user/project

Best for: Maximum control, debugging, understanding the resume process.

Command Mode (command) - Default

User explicitly runs the resume command with --exec flag.

$ egenskriven resume WRK-42 --exec
Resuming session for WRK-42...
Tool: opencode
Working directory: /home/user/project

[Agent session starts]

Best for: Most workflows, explicit control over when resume happens.

Auto Mode (auto)

Session resumes automatically when human adds a comment containing @agent.

# Human adds comment:
$ egenskriven comment WRK-42 "@agent I've decided to use JWT auth"

# Session automatically resumes with full context

Best for: Responsive workflows, quick back-and-forth communication.

Configuring Resume Mode

# View current mode
egenskriven board show <board>

# Change mode
egenskriven board update <board> --resume-mode auto

Workflow Recommendations

Collaborative Mode Integration

The resume modes work with all agent modes:

Example configuration:

{
  "workflow_mode": "light",
  "agent": {
    "mode": "collaborative"
  }
}

With collaborative mode, agents are expected to:

1. •Execute minor updates directly
2. •Block for major decisions requiring human input
3. •Provide clear, specific questions when blocking

Related Skills

• •egenskriven - Core commands and task management
• •egenskriven-advanced - Epics, dependencies, sub-tasks, batch operations