agile-workflow（敏捷的工作流程）

Agile Workflow Orchestrator

A skill that guides agents through structured agile development workflows by intelligently invoking commands in sequence. Uses checkpoint-based flow control to auto-progress between steps while pausing at key decision points.

Note: This is a platform-agnostic, git-only workflow. For PR-based workflows with specific platforms, use:

• gitea-workflow for Gitea repositories
• github-agile for GitHub repositories

When to Use This Skill

Use this skill when:

• Starting work for the day ("run morning standup", "start my day")
• Working on a task ("implement next task", "continue working")
• Completing a development cycle ("finish this task", "prepare PR")
• Running sprint ceremonies ("start sprint", "end sprint", "retrospective")
• Resuming interrupted work ("what's next", "where was I")

Do NOT use this skill when:

• Running a single specific command (use that command directly)
• Just checking status (use /status directly)
• Only doing code review without full cycle (use /review-code directly)
• Researching or planning without implementation

Prerequisites

Before using this skill:

• Git repository initialized with worktree support
• Context network with backlog structure at context-network/backlog/
• Task status files at context-network/backlog/by-status/*.md

Workflow Types Overview

WORKFLOW TYPES
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

TASK CYCLE (Primary)     DAILY                SPRINT
──────────────────────   ──────────────────   ──────────────────
sync                     Morning:             Start:
  ↓                        sync --last 1d       sync --all
next → [CHECKPOINT]        status --brief       groom --all
  ↓                        groom --ready        plan sprint-goals
implement                                       status
  ↓                      Evening:
[CHECKPOINT]               checklist          End:
  ↓                        discovery            sync --sprint
review-code                sync --last 1d       retrospective
review-tests                                    audit --sprint
  ↓                                             maintenance --deep
[CHECKPOINT]
  ↓
apply-recommendations (if issues)
  ↓
merge-prep → [CHECKPOINT]
  ↓
merge-complete
  ↓
update-backlog & status
  ↓
END
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

State Detection

The skill determines current workflow state automatically. No manual tracking needed.

Detection Signals

State Matrix

STATE DETECTION LOGIC
─────────────────────────────────────────────────────────────
Check                           → State           → Next Step
─────────────────────────────────────────────────────────────
No worktree, no in-progress     → IDLE            → sync, next
Worktree exists, uncommitted    → IMPLEMENTING    → continue implement
Worktree exists, all committed  → READY_REVIEW    → review-code
Reviews complete, ready to merge→ MERGE_READY     → merge-prep
Branch merged, worktree exists  → CLEANUP         → merge-complete
─────────────────────────────────────────────────────────────

For detailed detection algorithms, see references/state-detection.md.

Invocation Patterns

# Auto-detect state and continue from where you are
/agile-workflow

# Start specific workflow phase
/agile-workflow --phase task-cycle
/agile-workflow --phase daily-morning
/agile-workflow --phase daily-evening
/agile-workflow --phase sprint-start
/agile-workflow --phase sprint-end

# Resume work on specific task
/agile-workflow --task TASK-123

# Preview what would happen without executing
/agile-workflow --dry-run

Task Cycle Phase

The primary workflow for completing a single task from selection to merge.

Step 1: Sync Reality

Ensure context network matches actual project state.

Run: sync --last 1d --dry-run
Purpose: Detect drift between documented and actual state
Output: Sync report showing completions, partial work, divergences

Step 2: Select Task

Identify the next task to work on.

Run: next
Purpose: Find highest priority ready task
Output: Task ID, title, branch name suggestion

CHECKPOINT: TASK_SELECTED

• Pause to confirm task selection
• User can accept or choose different task
• On accept: continue to implementation

Step 3: Implement

Test-driven development in isolated worktree.

Run: implement [TASK-ID]
Purpose: Create worktree, write tests first, implement, verify
Output: Working implementation with passing tests

CHECKPOINT: IMPL_COMPLETE

• Pause after implementation completes
• Show test results and coverage
• On success: continue to review

Step 4: Review

Quality validation of implementation.

Run: review-code --uncommitted
Run: review-tests --uncommitted
Purpose: Identify quality issues, security concerns, test gaps
Output: Review reports with issues and recommendations

CHECKPOINT: REVIEWS_DONE

• Display combined review results
• If critical issues: must address before continuing
• If no issues: auto-continue to PR prep
• User decides: apply recommendations now or defer

Step 5: Apply Recommendations (Conditional)

Address review findings intelligently.

Run: apply-recommendations [review-output]
Purpose: Apply quick fixes now, defer complex changes to tasks
Output: Applied fixes + created follow-up tasks

Step 6: Prepare Merge

Validate and prepare for merge to main.

Run: merge-prep
Purpose: Validate implementation, run final checks, prepare for merge
Output: Ready to merge to main

CHECKPOINT: MERGE_READY

• Display validation results
• Show files that will be merged
• On all checks pass: continue to merge
• On failure: stop, address issues

Step 7: Complete Merge

Merge to main and cleanup.

Run: merge-complete
Purpose: Merge to main, delete branch, remove worktree, update status
Output: Task marked complete, cleanup done

Step 8: Update Backlog and Project Status

Persist progress to source-of-truth documentation.

Run: Part of merge-complete (Phase 6)
Purpose: Update epic file (task → complete), unblock dependents, update project status
Output: Backlog and project status reflect actual progress

Why this step matters: Without it, completed tasks remain marked "ready" in backlog files and project status stays stale. Internal tracking files are session-scoped; the backlog and status files are the persistent source of truth.

For detailed task-cycle instructions, see references/phases/task-cycle.md.

Daily Phase

Quick sequences for start and end of workday.

Morning Standup (~5 min)

Run sequence:
1. sync --last 1d --dry-run   # What actually happened yesterday
2. status --brief --sprint    # Current sprint health
3. groom --ready-only         # What's ready to work on

Output: Clear picture of today's priorities

Evening Wrap-up (~10 min)

Run sequence:
1. checklist                  # Ensure nothing lost
2. discovery                  # Capture learnings
3. sync --last 1d            # Update task statuses

Output: Knowledge preserved, state synchronized

For detailed daily instructions, see references/phases/daily.md.

Sprint Phase

Ceremonies for sprint boundaries.

Sprint Start (~60 min)

Run sequence:
1. sync --all                 # Full reality alignment
2. groom --all               # Comprehensive grooming
3. plan sprint-goals         # Architecture and goals
4. status --detailed         # Baseline metrics

Output: Sprint plan with groomed, ready backlog

Sprint End (~90 min)

Run sequence:
1. sync --sprint             # Final sprint sync
2. retrospective             # Capture learnings
3. audit --scope sprint      # Quality review
4. status --metrics          # Sprint metrics
5. maintenance --deep        # Context network cleanup

Output: Sprint closed, learnings captured, ready for next

For detailed sprint instructions, see references/phases/sprint.md.

Checkpoint Handling

Checkpoints are pauses for human decision-making.

Checkpoint Behavior

At each checkpoint:

1. Summarize what just completed
2. Show key results and any issues
3. Present next steps
4. Wait for user input

Checkpoint Responses

Auto-Continue Conditions

Some checkpoints can auto-continue when conditions are met:

For detailed checkpoint handling, see references/checkpoint-handling.md.

Command Reference

Each workflow step uses embedded command instructions:

Example: Complete Task Cycle

Scenario: Start of day, implement next available task

Invocation:

/agile-workflow --phase task-cycle

Flow:

Agent: Detecting current state...
       No worktree found, checking ready tasks...

Agent: Running sync --last 1d --dry-run
       [Sync output: 0 tasks completed but undocumented]

Agent: Running next
       Found: TASK-042 - Add user preferences endpoint
       Priority: High | Size: Small | Branch: task/TASK-042-preferences

       ╔═══════════════════════════════════════════╗
       ║  CHECKPOINT: Task Selection               ║
       ╠═══════════════════════════════════════════╣
       ║  Selected: TASK-042                       ║
       ║  Proceed with implementation?             ║
       ║                                           ║
       ║  [continue] [stop] [other task]           ║
       ╚═══════════════════════════════════════════╝

User: continue

Agent: Creating worktree at .worktrees/TASK-042/
       Running implement TASK-042
       [Implementation proceeds with TDD...]

       Tests: 12 passing
       Coverage: 87%
       Build: Success

       ╔═══════════════════════════════════════════╗
       ║  CHECKPOINT: Implementation Complete      ║
       ╠═══════════════════════════════════════════╣
       ║  All tests passing, ready for review      ║
       ║  Proceed to code review?                  ║
       ║                                           ║
       ║  [continue] [stop]                        ║
       ╚═══════════════════════════════════════════╝

[Flow continues through review, merge...]

Limitations

• Requires context network with specific backlog structure
• Git-only workflow (no PR integration - for PR workflows use gitea-workflow or github-agile)
• Single-task focus (parallel task work not orchestrated)
• Best suited for solo development or trusted team direct-to-main workflows

Related Skills

• gitea-workflow - For Gitea repositories with PR integration
• github-agile - For GitHub repositories with PR integration
• skill-maker - Create new skills following agentskills.io spec
• research-workflow - For research tasks before implementation