Todo Manager Skill

Overview

This skill provides a complete task management system for AI agents. It enables autonomous task execution, scheduled work sessions, git integration, and quality verification. The skill operates in 5 distinct modes to handle all aspects of project task management.

When to Use

Activate this skill when the user mentions:

• "start a todo list" or "task list"
• "task management" or "project tracking"
• "work on tasks" or "execute tasks"
• "automated task execution" or "autonomous work"
• "scheduled tasks" or "recurring work"
• Any request to create or manage project tasks

File Formats

.todo.md Structure

The task file uses a structured markdown format with task blocks:

markdown
---
title: Todo List
project: [Project Name]
created: [ISO timestamp]
last_updated: [ISO timestamp]
scheduled_task_id: [UUID from scheduler:create_scheduled_task]
scheduled_task_mode: work | verify | complete
work_complete: true | false
---

## Task 1: [Task Title]
id: [auto-generated unique task ID, e.g., task-001]
- **Status:** not started | in progress | complete | failed | verified
- **Created:** [ISO timestamp]
- **Updated:** [ISO timestamp]
- **Description:** [What needs to be done - be specific and detailed]
- **Notes:** [Work notes, added as tasks progress]
- **Subtasks:** [Optional list of subtasks for this task]

## Task 2: [Task Title]
id: [auto-generated unique task ID, e.g., task-002]
- **Status:** not started | in progress | complete | failed | verified
- **Created:** [ISO timestamp]
- **Updated:** [ISO timestamp]
- **Description:** [What needs to be done]
- **Notes:** [Work notes]
- **Subtasks:** [Optional list of subtasks for this task]

CHANGELOG.md Format (Common Changelog Standard)

markdown
# Changelog

All notable changes to this project will be documented in this file.

The format is based on [Common Changelog](https://commonform.github.io/changelog/).

## [Date] - Task: [Task Title]

### Completed
- [Description of work completed]

### Notes
- [Additional notes about the work]

---

## [Previous entries...]

error.md Format

markdown
# Todo Manager Error

- **Error:** [Error description]
- **Timestamp:** [ISO timestamp]
- **Context:** [What the agent was trying to do]
- **Resolution:** [What action was taken]

Mode Definitions

MODE 1: INIT_MODE (Initialization)

Purpose: Set up new task management for a project.

Entry Condition: First time user wants to start task management OR explicitly requested.

CRITICAL: Tasks are NEVER deleted - Only status changes throughout the lifecycle.

Actions:

1. Check if .todo.md already exists in project root
2. If exists, present current state to user and ask:

> "A task list already exists for this project. What would you like to do? > > Options: > - Replace: Clear all existing tasks and start fresh with new goals > - Modify: Keep existing tasks and add/update as needed > > Current task list: > List all existing tasks with their status]

1. If user chooses Replace:

• Clear all tasks from .todo.md (keep header metadata)
• Ask user for new task list
• Proceed to populate with new tasks

1. If user chooses Modify:

• Walk through current task list with user
• Ask which tasks to keep, update, or remove
• Ask what new goals/tasks to add
• Update .todo.md accordingly

1. If not exists, create .todo.md with project header
2. Create CHANGELOG.md if it doesn't exist
3. Ask user for task list (what needs to be done)
4. Populate .todo.md with tasks
5. Ask user for work frequency (how often to work on tasks)
6. Schedule recurring task using scheduler:create_scheduled_task

• CRITICAL: Save the returned UUID as scheduled_task_id in frontmatter
• Set scheduled_task_mode: work and work_complete: false in frontmatter
• Write the updated frontmatter to .todo.md

1. Confirm setup complete with user

MODE 2: WORK_MODE (Task Execution)

Purpose: Execute the next task in the queue autonomously.

Entry Condition:

• Scheduled task fires with scheduled_task_mode: work in frontmatter, OR
• User says "work on tasks" / "execute tasks"

CRITICAL: When to Enter VERIFY_MODE: After completing a task, ALWAYS check if ALL tasks are now complete. If all tasks have status "complete" or "verified", you MUST:

1. Update the .todo.md frontmatter to change scheduled_task_mode: verify
2. Then enter VERIFY_MODE immediately to verify the completed work

CRITICAL: Tasks are NEVER deleted - Only status changes throughout the lifecycle.

Actions:

1. Check if .todo.md exists

• If not → enter ERROR_MODE

1. Check git status with git status

• If dirty → enter CORRECTION_MODE

1. Read scheduled_task_mode and work_complete from frontmatter

• If work_complete is "true" → All work already done, notify user
• If scheduled_task_mode is "verify" → enter VERIFY_MODE
• If scheduled_task_mode is "complete" → All work done, notify user

1. Find first task with status "not started"

• If none → Check if any tasks have status "complete"
• If yes → ALL WORK IS DONE, update frontmatter to scheduled_task_mode: verify, enter VERIFY_MODE
• If no tasks are "complete" either → Notify user all tasks are done

1. Set task status to "in progress", add timestamp to notes
2. Work the task autonomously using sub-agents/subordinates - break down the task into subtasks and delegate to subordinate agents using call_subordinate, solve issues without waiting for input

• Subtask limit: Work on a maximum of 3 subtasks per session
• If task has more than 3 subtasks, complete up to 3 and mark remaining for next session
• Help user break down large tasks into smaller, manageable chunks during INIT_MODE

1. If task is too large to complete in one session:

• Break down the task into smaller subtasks
• Add new tasks to .todo.md with unique IDs
• Mark current task as "in progress" with notes on what's done
• Commit progress so far

1. On completion:

• Set status to "complete"
• Add completion notes to task
• Update CHANGELOG.md with descriptive notes
• Commit all work with single commit
• CRITICAL: After marking complete, check if ALL tasks are now complete or verified
• If ALL tasks are complete/verified:
• Update .todo.md frontmatter: scheduled_task_mode: verify
• Write the updated frontmatter to .todo.md
• Enter VERIFY_MODE immediately to verify completed work

1. On failure:

• Set status to "failed"
• Add detailed failure notes explaining why
• Create new investigation task placed directly before failing task
• Include context from failure
• Commit all work with single commit

MODE 3: VERIFY_MODE (Quality Review)

Purpose: Review completed tasks to verify quality. When all tasks are verified, disable the scheduled task so it no longer runs. Tasks are NEVER deleted - only status changes.

Entry Condition:

• Scheduled task fires with scheduled_task_mode: verify in frontmatter, OR
• WORK_MODE completes last "not started" task and all tasks are now "complete" or "verified"

CRITICAL: When to Enter VERIFY_MODE from WORK_MODE: The most important trigger is when WORK_MODE completes a task and discovers there are no more "not started" tasks. This means all work is done and verification is needed.

CRITICAL: Tasks are NEVER deleted - Only status changes throughout the lifecycle.

Actions:

1. Check if there are any tasks with status "complete" (not yet verified)

• If none → ALL TASKS ALREADY VERIFIED, skip to step 6

1. Get the first task marked "complete" (not yet "verified")
2. Review the task description and completion notes
3. Verify requirements were met and work is quality

• Check if all subtasks were addressed
• Verify code/tests work if applicable
• Confirm documentation is complete

1. If verification passes:

• Change status from "complete" to "verified"
• Add verification notes to task
• Update CHANGELOG.md with verification notes
• Commit all changes
• Loop back to step 1 to check for more "complete" tasks

1. If verification fails:

• Set status back to "in progress"
• Add notes about what needs to be fixed
• Update frontmatter: scheduled_task_mode: work (switch back to work mode)
• Write updated frontmatter to .todo.md
• Return to WORK_MODE

1. After all tasks verified (no "complete" tasks remain, only "verified"):

• Read the scheduled_task_id from .todo.md frontmatter
• DISABLE the scheduled task using scheduler:update_task:
• Tool: scheduler:update_task
• Arguments: task_id: scheduled_task_id from frontmatter], state: "disabled"
• Update .todo.md frontmatter to indicate work is complete:

yaml scheduled_task_mode: complete work_complete: true

• Write updated frontmatter to .todo.md
• Notify user that all tasks are complete and scheduled task has been disabled

MODE 4: CORRECTION_MODE (Git State Recovery)

Purpose: Handle dirty git state and get back on track.

Entry Condition: Git status shows uncommitted changes when entering WORK_MODE

Actions:

1. Analyze the uncommitted changes
2. Read .todo.md to find the last task that was being worked on
3. Determine if the task was completed:

• Check if task status was set to "complete" before git state became dirty
• Check commit history for recent task completion commits

1. If task was NOT completed:

• Set task status back to "in progress" if needed
• Continue working on the task in WORK_MODE

1. If task was completed (status already "complete"):

• Commit the changes with descriptive message
• Return to normal WORK_MODE to find next task

1. If unclear:

• Add notes to current task about the situation
• Commit with message: "WIP: task name] - state recovery"
• Return to WORK_MODE

MODE 5: ERROR_MODE (Error Handling)

Purpose: Handle errors when task file is missing.

Entry Condition: .todo.md does not exist and not in INIT_MODE

Actions:

1. Create error.md with:

• Error: "Project does not contain .todo.md"
• Timestamp
• Context: What the agent was trying to do
• Resolution: "Created error.md - INIT_MODE required"

1. Check if error.md already exists with same information

• If same error exists:
• Use scheduler:find_task_by_name to find recurring work task
• Disable the recurring task using scheduler:update_task with state: "disabled"
• Notify user that task management was cancelled
• If new error:
• Ask user if they want to start task management (enter INIT_MODE)

Step-by-Step Instructions

Starting Task Management (INIT_MODE)

1. Check existing files:

bash ls -la | grep -E "\.todo\.md|CHANGELOG\.md|error\.md"

1. If .todo.md exists, present current state and ask user:

> "A task list already exists for this project. Here's the current state: > > Current Tasks: > - Task 1: Title] - Status] > - Task 2: Title] - Status] > - ... > > What would you like to do? > - Replace: Clear all tasks and start fresh with new goals > - Modify: Update the existing list (keep, remove, or change tasks)"

1. If user chooses Replace:

• Ask: "Are you sure you want to replace the current task list? This will clear all existing tasks."
• If confirmed, clear tasks from .todo.md (keep header metadata)
• Ask for new task list

1. If user chooses Modify:

• Walk through current tasks one by one:
• For each task: "Keep this task? Update it? Remove it?"
• Ask: "What new tasks would you like to add?"
• Update .todo.md with changes

1. If no files, create .todo.md with ALL required fields:

markdown --- title: Todo List project: [Project Name] created: [ISO timestamp] last_updated: [ISO timestamp] scheduled_task_id: "" # Will be filled after creating scheduled task scheduled_task_mode: work work_complete: false ---

1. Create CHANGELOG.md:

markdown # Changelog

All notable changes to this project will be documented in this file.

The format is based on Common Changelog.

---

1. Ask user for tasks:

> "What tasks need to be completed for this project? Please describe each task clearly."

1. Populate tasks in .todo.md with status "not started"

1. Ask for frequency with recommended options:

> "How often should I work on these tasks? Here are some recommended options: > > Recommended Intervals: > - 15 minutes - Quick check-ins, small tasks > - 20 minutes - Small to medium tasks > - 30 minutes - Medium tasks (recommended default) > - 45 minutes - Medium to large tasks > - 60 minutes - Large, complex tasks > > Note: If your tasks are large or complex, I recommend a larger frequency (30-60 minutes) to prevent agent overlap and ensure each session can complete meaningful work before the next session starts. > > How often would you like me to work on these tasks?"

1. Create scheduled task:

• Use scheduler:create_scheduled_task with:
• Name: "Project] Todo Work Session"
• Prompt: Instructions to load todo-manager skill and execute appropriate mode based on scheduled_task_mode
• Schedule based on user frequency
• dedicated_context: true (so it runs in its own context)
• CRITICAL: Get the returned UUID from scheduler:create_scheduled_task
• Update .todo.md frontmatter with the UUID:

yaml scheduled_task_id: "[UUID from scheduler response]" scheduled_task_mode: work work_complete: false

• Write the complete updated .todo.md file with the scheduled_task_id

Working Tasks (WORK_MODE)

CRITICAL: When to Enter VERIFY_MODE After EVERY task completion, you MUST check if all tasks are now complete or verified. This is the key trigger for VERIFY_MODE.

CRITICAL: Tasks are NEVER deleted - Only status changes.

1. Check for .todo.md:

bash test -f .todo.md && echo "exists" || echo "missing" If missing → ERROR_MODE

1. Check git status:

bash git status --porcelain If output not empty → CORRECTION_MODE

1. Read frontmatter from .todo.md:

• Parse .todo.md to get scheduled_task_mode and work_complete values
• If work_complete is "true" → All work done, notify user and exit
• If scheduled_task_mode is "verify" → enter VERIFY_MODE
• If scheduled_task_mode is "complete" → All work done, notify user and exit

1. Find next task:

Parse .todo.md looking for first task with status "not started"

1. If no "not started" tasks remain:

• Check if any tasks have status "complete" (not yet verified)
• If yes → ALL WORK IS DONE!
• Update .todo.md frontmatter: scheduled_task_mode: verify
• Write updated frontmatter to .todo.md
• Enter VERIFY_MODE immediately to verify completed work
• If no "complete" tasks either → All tasks already verified, enter VERIFY_MODE to complete cleanup

1. Update task to "in progress":

markdown

• Status: in progress
• Updated: ISO timestamp]
• Notes: Started work at timestamp]

1. Execute task autonomously using sub-agents/subordinates - DO NOT wait for user input

• Break down task into subtasks
• Delegate subtasks to subordinate agents using call_subordinate tool
• Work on maximum 3 subtasks per session - if more exist, complete up to 3 and note the rest for next session
• Execute each subtask via sub-agents
• Handle errors and issues independently
• Use all available tools to solve problems

1. If task is too large:

• Break down into smaller tasks
• Add new tasks to .todo.md with unique IDs
• Mark current task as partially complete with progress notes
• Commit progress

1. On completion:

markdown

• Status: complete
• Updated: ISO timestamp]
• Notes: What was accomplished]

Update CHANGELOG.md: markdown ## Date] - Task: Task Title]

### Completed

• Description of work]

### Notes

• Any additional notes]

Commit: bash git add -A git commit -m "Complete: [Task Title] - [Brief description]"

CRITICAL: Check if ALL tasks are now complete:

• Parse .todo.md for all task statuses
• If ALL tasks have status "complete" or "verified" (none are "not started" or "in progress" or "failed"):
• Update frontmatter: scheduled_task_mode: verify
• Write updated frontmatter to .todo.md
• Enter VERIFY_MODE immediately to verify the completed work

1. On failure:

markdown

• Status: failed
• Updated: ISO timestamp]
• Notes: Detailed explanation of why it failed, error messages, attempts made]

Create investigation task directly BEFORE the failed task: markdown ## Task X: Investigate: Failed Task Title] Failure

• Status: not started
• Created: ISO timestamp]
• Description: Investigate and resolve the failure: failure context]
• Notes: Context from failed task]

Commit: bash git add -A git commit -m "Failed: [Task Title] - Reason: [brief reason]. Created investigation task."

Verifying Completed Tasks (VERIFY_MODE)

When to enter VERIFY_MODE:

• Scheduled task fires with scheduled_task_mode: verify in frontmatter, OR
• WORK_MODE completes last "not started" task and all remaining tasks are "complete"

NOTE: VERIFY_MODE may be entered multiple times - it verifies ONE task per run, then the next scheduled run will verify the next task. This is by design to prevent one agent from doing too much work.

CRITICAL: Tasks are NEVER deleted - Only status changes.

1. Check for remaining "complete" tasks:

• Parse .todo.md looking for tasks with status "complete"
• If none found → All tasks already verified, skip to step 6

1. Get first "complete" task from .todo.md (first task with status "complete", not yet "verified")

1. Verify the work:

• Re-read task description
• Check if requirements were met
• Test code if applicable
• Review documentation

1. If verified:

• Change status from "complete" to "verified"
• Add verification notes:

markdown

• Status: verified
• Verified: ISO timestamp]
• Verification Notes: Quality check passed - notes]

• Update CHANGELOG.md:

markdown ## Date] - Task: Task Title] - VERIFIED

### Completed

• Original completion notes]

### Verified

• Quality check passed - verification notes]

• Commit:

bash git add -A git commit -m "Verified: [Task Title] - Quality verified"

• The scheduled task will run again to verify the next "complete" task

1. If failed verification:

• Change status back to "in progress"
• Add notes about what needs fixing
• Update frontmatter: scheduled_task_mode: work (switch back to work mode)
• Write updated frontmatter to .todo.md
• Continue in WORK_MODE

1. After all tasks verified (no "complete" tasks remain):

• DISABLE the scheduled task using scheduler:update_task:
• Tool: scheduler:update_task
• Arguments:
• task_id: scheduled_task_id from frontmatter]
• state: "disabled"
• Update .todo.md frontmatter to indicate completion:

yaml scheduled_task_mode: complete work_complete: true

• Write updated frontmatter to .todo.md
• Notify user that all tasks are complete and scheduled task has been disabled

Handling Dirty Git State (CORRECTION_MODE)

1. Get git diff to see what changed:

bash git diff --stat git diff

1. Read .todo.md to see last task status

1. Check commit history:

bash git log --oneline -10

1. Determine action:

• If task was marked "complete" before dirty state → commit and continue
• If task was in progress → continue working
• If unclear → commit with "WIP" message and continue

Error Handling (ERROR_MODE)

1. Check if error.md exists:

bash test -f error.md && cat error.md

1. If same error exists:

bash scheduler:find_task_by_name # Find the recurring work task scheduler:update_task (if found)

1. If new error:

Create error.md with details Ask user: "This project doesn't have task management set up. Would you like to start a new task list?"

Frequency Options

Recommended work frequencies for scheduled tasks:

| Interval | Best For | Notes | |----------|----------|-------| | 15 minutes | Quick check-ins, very small tasks | May cause overlap with large tasks | | 20 minutes | Small tasks | Minimum for most sessions | | 30 minutes | Medium tasks (recommended default) | Good balance | | 45 minutes | Medium to large tasks | Recommended for complex work | | 60 minutes | Large, complex tasks | Prevents agent overlap |

Guidance for Large Tasks:

If your task list contains large or complex tasks, choose a larger frequency (30-60 minutes). This ensures:

• Each session has enough time to make meaningful progress
• Previous session completes before next starts (no agent overlap)
• Sub-agents have sufficient time to complete subtasks
• Better git commit hygiene (one task per session)

| User Input | Cron Schedule | |------------|---------------| | Every 15 minutes | */15 * * * * | | Every 20 minutes | */20 * * * * | | Every 30 minutes | */30 * * * * | | Every 45 minutes | */45 * * * * | | Every 60 minutes | 0 * * * * | | Hourly | 0 * * * * | | Daily | 0 0 * * * | | Weekly | 0 0 * * 0 |

Or accept custom cron expression from user.

Available Scheduler Tools

IMPORTANT: Use these exact tool names and parameters:

| Tool | Purpose | Key Parameters | |------|---------|-----------------| | scheduler:create_scheduled_task | Create recurring task | name, system_prompt, prompt, schedule, dedicated_context | | scheduler:update_task | Update or disable scheduled task | task_id, state (use "disabled" to disable) | | scheduler:find_task_by_name | Find task by name | name | | scheduler:run_task | Run task manually | uuid, context | | scheduler:delete_task | Delete scheduled task | uuid (use update_task with state: "disabled" instead when possible) |

Example Use Cases

Example 1: Starting a New Project

• User: "I want to start a todo list for my project"
• Agent: Enters INIT_MODE, checks for existing files
• Result: .todo.md, CHANGELOG.md created with scheduled_task_id, scheduled task configured

Example 2: Continuing Existing Project with New Goals

• User: "I want to add more tasks"
• Agent: Enters INIT_MODE, finds existing .todo.md
• Agent: Presents current tasks, asks: "Replace or Modify?"
• User: "Modify"
• Agent: Walks through current tasks, adds new ones
• Result: Updated .todo.md with new and existing tasks

Example 3: Scheduled Work Session - All Work Complete

• Trigger: Scheduled task fires, agent enters WORK_MODE
• Agent: Checks scheduled_task_mode from frontmatter
• Agent: Finds no "not started" tasks but has "complete" tasks
• Agent: Updates frontmatter to scheduled_task_mode: verify, writes to file, enters VERIFY_MODE
• Result: Completed task is verified, status updated to "verified"

Example 4: Verify Mode - Next Run Verifies Next Task

• Trigger: Scheduled task fires with scheduled_task_mode: verify
• Agent: Enters VERIFY_MODE, finds first "complete" task
• Agent: Verifies it, updates status to "verified", commits
• Agent: Checks for more "complete" tasks - if more exist, the next scheduled run will verify the next one
• Result: One task verified per run, eventually all verified

Example 5: All Work Complete - Disable Scheduled Task

• Trigger: VERIFY_MODE finds no remaining "complete" tasks
• Agent: Reads scheduled_task_id, calls scheduler:update_task with state: "disabled"
• Agent: Updates frontmatter to scheduled_task_mode: complete, work_complete: true
• Result: Project complete, scheduled task disabled, .todo.md preserved with all tasks (status changed but never deleted)

Example 6: User Requests Work

• User: "Work on the tasks"
• Agent: Enters WORK_MODE (checks git first), executes task
• Result: Task completed or failed with investigation task created

Example 7: Dirty Git State

• Trigger: WORK_MODE finds dirty git
• Agent: Enters CORRECTION_MODE, analyzes state
• Result: Either commits and continues OR continues interrupted work

Example 8: No Todo File

• Trigger: Agent in non-INIT mode finds no .todo.md
• Agent: Enters ERROR_MODE, creates error.md
• Result: User prompted to start task management or recurring task disabled

Edge Cases

1. No tasks in .todo.md

• Ask user to add tasks before scheduling

1. All tasks failed

• Keep failed tasks with investigation tasks
• Agent will work on investigation tasks first

1. Recurring task already exists

• Ask user if they want to update frequency or keep existing

1. Git conflict during commit

• Resolve conflict, complete commit, note in changelog

1. Very long task descriptions

• Keep descriptions concise but complete
• Add detail in subtasks if needed

1. Task depends on another task

• Add dependency notes in description
• Agent should complete prerequisite tasks first

1. Large task frequency recommendation

• Always recommend 30-60 minute intervals for large/complex tasks
• Explain that smaller intervals may cause agent overlap

1. Verify mode not triggering

• CRITICAL: After completing ANY task in WORK_MODE, always check if all tasks are now complete
• If all tasks are "complete" or "verified", immediately update scheduled_task_mode: verify and enter VERIFY_MODE
• This is the key mechanism that ensures verification happens

1. Scheduled task runs in wrong mode

• The frontmatter scheduled_task_mode field controls which mode runs
• WORK_MODE updates this to "verify" when all tasks complete
• VERIFY_MODE updates this to "work" if verification fails and work is needed

1. Task verification takes multiple runs

• VERIFY_MODE verifies ONE task per scheduled run
• This is intentional to prevent agent overload
• The next scheduled run will verify the next "complete" task

Fundamental Principles

CRITICAL: Tasks are NEVER deleted

• Tasks remain in .todo.md forever
• Only status changes: not started → in progress → complete → verified (or failed)
• This preserves the complete history of work done

CRITICAL: Scheduled Task Handling

• The recurring scheduled task should be DISABLED when all work is complete
• Use scheduler:update_task with state: "disabled" to disable the task
• The .todo.md file itself is NEVER deleted

Scheduled Task Prompt Template

When creating the recurring task, use this prompt:

You are running a scheduled work session for the todo-manager skill.

Project: [PROJECT NAME]
Task File: .todo.md

IMPORTANT: Check the scheduled_task_mode and work_complete in .todo.md frontmatter FIRST:
- If work_complete is "true": All work is done, notify user and exit
- If scheduled_task_mode is "complete": All work is done, notify user and exit
- If scheduled_task_mode is "verify": Execute VERIFY_MODE
- If scheduled_task_mode is "work" OR NOT SET: Execute WORK_MODE

Instructions:
1. Load the todo-manager skill
2. Read .todo.md and parse the frontmatter for scheduled_task_mode and work_complete
3. Execute the appropriate mode based on the mode values
4. Do not wait for user input - work autonomously
5. Handle any errors that arise
6. Report completion status when done

Remember: Use all available tools to complete tasks. Do not ask for user input unless absolutely necessary.

Tools & Resources

Required Tools

• code_execution_tool: For git operations, file read/write
• scheduler:create_scheduled_task: For creating scheduled tasks (returns UUID)
• scheduler:update_task: For disabling scheduled task (use state: "disabled")
• scheduler:find_task_by_name: For finding existing tasks
• document_query: For reading task files
• memory_*: For storing project context if needed

File Locations

• Task file: .todo.md (project root) - NEVER deleted
• Changelog: CHANGELOG.md (project root)
• Error file: error.md (project root)

Notes

• Always use ISO 8601 timestamps: YYYY-MM-DDTHH:MM:SSZ
• The skill assumes it's running in the project directory
• Commits should be atomic - one task per commit
• Use descriptive commit messages that match changelog entries
• Investigation tasks should have clear failure context
• When in doubt, prefer continuing work over asking questions
• Check git status before every work session
• Quality verification is critical - don't just mark complete
• CRITICAL: When existing tasks are found, ALWAYS ask user whether to replace or modify - never automatically reset
• CRITICAL: Recommend larger frequencies (30-60 min) for large tasks to prevent agent overlap
• CRITICAL: After completing ANY task in WORK_MODE, ALWAYS check if ALL tasks are now complete/verified. If so, update scheduled_task_mode: verify, write to file, and enter VERIFY_MODE immediately.
• CRITICAL: In VERIFY_MODE, when a task passes verification, update its status to "verified" (NOT deleted)
• CRITICAL: Always save the scheduled_task_id from scheduler:create_scheduled_task to the frontmatter - this is required to disable the task later
• CRITICAL: Tasks are NEVER deleted from .todo.md - only status changes
• CRITICAL: Use scheduler:update_task with state: "disabled" to disable the scheduled task when work is complete

Skill Version: 1.5 | Author: Agent Zero | Based on agentskills.io specification

Version History

• 1.5 - Fixed: scheduler:update_task DOES exist with state parameter. Use scheduler:update_task with state: "disabled" to disable scheduled task (not delete). Added work_complete check at start of WORK_MODE.
• 1.4 - Clarified: tasks NEVER deleted (only status changes), scheduled task removed using scheduler:delete_task when complete, added work_complete flag to frontmatter
• 1.3 - Fixed critical bugs: scheduler:update_task replaced with scheduler:delete_task, added scheduled_task_id saving in INIT_MODE, fixed VERIFY_MODE to verify one task per run with loop back, added scheduled_task_mode check at start of WORK_MODE
• 1.2 - Added VERIFY_MODE trigger mechanism: WORK_MODE now checks if all tasks complete after each task, updates scheduled_task_mode to "verify", and enters VERIFY_MODE immediately. Verified tasks now marked as "verified" status.
• 1.1 - Added replace/modify workflow for existing todo lists and frequency recommendations (15/20/30/45/60 min) with guidance for large tasks
• 1.0 - Initial release