Checkpoint Manager Skill

A skill for consistently tracking and reporting task progress.

Auto-Suggestion Triggers

Situation	Suggestion
Conversation start	"Shall I check for pending tasks?"
New task request	"Shall I start a checkpoint?"
Before loading instructions	"Shall I record instruction usage?"
Work milestone	"Shall I report progress?"
Work completion	"Shall I complete the task?"

Workflow Overview

code

┌─────────────┐    ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  pending    │    │  start      │    │  progress   │    │  complete   │
│  check      │ → │  begin      │ → │  report     │ → │  finish     │
└─────────────┘    └─────────────┘    └─────────────┘    └─────────────┘
                          ↓
                   ┌─────────────┐
                   │ instruction │
                   │ management  │
                   └─────────────┘

See workflow.md for details.

Command Reference

Task Management

bash

# Check pending tasks
scripts/checkpoint.sh pending

# Start a new task (task ID auto-generated)
scripts/checkpoint.sh start "<task-name>" <steps>

# Report progress (only during instruction use)
scripts/checkpoint.sh progress <task-id> <current> <total> "<status>" "<next>"

# Complete task (all instructions must be completed)
scripts/checkpoint.sh complete <task-id> "<result>"

# Report error
scripts/checkpoint.sh error <task-id> "<message>"

Instruction Management

bash

# Start using an instruction
scripts/checkpoint.sh instruction-start "<path>" "<purpose>" [task-id]

# Complete instruction use
scripts/checkpoint.sh instruction-complete "<path>" "<result>" [task-id]

Status Check

bash

# Show task history
scripts/checkpoint.sh summary <task-id>

Usage Scenarios

Scenario 1: Conversation Start

When user starts a new conversation, suggest checking for pending tasks:

code

AI: Shall I check for pending tasks?

# Execute
scripts/checkpoint.sh pending

Scenario 2: New Task Request

When user requests something like "implement X":

code

AI: Shall I start a checkpoint?
    Task name: [user's request]
    Estimated steps: [based on complexity]

# Execute
scripts/checkpoint.sh start "Feature implementation" 5
# → Task ID: TASK-123456-abc123

Scenario 3: Instruction Use

Before loading an instruction:

code

AI: Shall I record instruction usage?

# Execute
scripts/checkpoint.sh instruction-start "instructions/en/presets/web_api.md" "REST API development" TASK-123456-abc123

Scenario 4: Work Milestone

When significant progress is made:

code

AI: Shall I report progress?
    Current: 2/5 steps
    Status: Design complete
    Next: Start implementation

# Execute
scripts/checkpoint.sh progress TASK-123456-abc123 2 5 "Design complete" "Start implementation"

Scenario 5: Work Completion

When task is complete:

code

AI: Shall I complete the task?
    Result: [summary of work]

# Execute instruction completion first
scripts/checkpoint.sh instruction-complete "instructions/en/presets/web_api.md" "3 endpoints implemented" TASK-123456-abc123

# Then complete task
scripts/checkpoint.sh complete TASK-123456-abc123 "REST API 3 endpoints implemented"

Workflow Constraints

Constraint	Description
Progress report	Only allowed during instruction use
Task completion	All instructions must be completed first
Task ID omission	Warning displayed for instruction commands

Decision Criteria

When to Suggest Task Start

•User requests specific work
•Work involves multiple steps
•Keywords like "implement", "create", "fix"

When to Suggest Progress Report

•Significant code changes made
•Tests passed/failed
•Clear work milestone reached

When to Suggest Task Completion

•User says "done", "finished", "complete"
•All requirements met
•Tests passing

Notes

•Task IDs are auto-generated (e.g., TASK-123456-abc123)
•Keep status and actions brief and clear
•Use the same task ID throughout a task
•Run scripts/checkpoint.sh from project root