AgentSkillsCN

feature-continue

从 task-list.md 中恢复被中断的功能开发，并通过 PM-DB 进行全程跟踪。

SKILL.md
--- frontmatter
name: feature-continue
description: Resume interrupted feature work from task-list.md with PM-DB tracking
args:
  task_list_path:
    type: string
    description: Path to task-list.md file
    required: true

feature-continue Skill

Resume interrupted feature development with PM-DB tracking intact.

Usage

bash
/feature-continue ./job-queue/feature-auth/tasks.md
/feature-continue ./my-feature/tasks.md

What It Does

Intelligently resumes feature work by:

  1. •Detecting where you left off
  2. •Checking PM-DB for existing phase_run
  3. •Resuming execution from the last incomplete task
  4. •Maintaining PM-DB tracking continuity

Workflow

Step 1: Detect Feature Location

code
🔍 Analyzing feature location...

Task list: ./job-queue/feature-auth/tasks.md
Input folder: ./job-queue/feature-auth
Planning folder: ./job-queue/feature-auth/planning

Step 2: Check PM-DB Status

code
🔍 Checking PM-DB tracking...

Query: SELECT * FROM phase_runs WHERE plan_id = (
  SELECT id FROM phase_plans WHERE phase_id = (
    SELECT id FROM phases WHERE name = 'feature-auth'
  )
) ORDER BY created_at DESC LIMIT 1;

Possible outcomes:

A) Active phase_run found:

code
✅ Found active phase run
   Phase Run ID: 78
   Phase ID: 34
   Plan ID: 56
   Status: in_progress
   Started: 2026-01-30 14:23:15

   Resuming from phase run #78...

B) Completed phase_run found:

code
⚠️ Found completed phase run
   Phase Run ID: 78
   Status: completed
   Completed: 2026-01-30 16:45:32

   This phase is already complete.

   Options:
   1. View metrics: /pm-db dashboard
   2. Start new run: /start-phase execute {path}
   3. Cancel

C) No phase_run found:

code
⚠️ No PM-DB tracking found

   This feature hasn't been started with PM-DB tracking.

   Options:
   1. Import to PM-DB: /pm-db import
   2. Start fresh: /feature-new "{description}"
   3. Execute without PM-DB: /start-phase execute {path}

Step 3: Detect Last Completed Task

code
🔍 Detecting progress...

Query: SELECT task_key FROM task_runs WHERE phase_run_id = 78 AND exit_code = 0;

Output:

code
✅ Found completed tasks:
   [1/7] Task 1: Setup auth routes ✅
   [2/7] Task 2: Create user model ✅
   [3/7] Task 3: Implement JWT ✅
   [4/7] Task 4: Add login endpoint ⏸️ (incomplete)
   [5/7] Task 5: Add logout endpoint ⏹️ (not started)
   [6/7] Task 6: Add protected routes ⏹️ (not started)
   [7/7] Task 7: Write tests ⏹️ (not started)

Last completed: Task 3
Next task: Task 4

Step 4: Resume Execution

code
🚀 Resuming phase execution...

Phase Run ID: 78
Starting from: Task 4
Remaining tasks: 4

Calling /start-phase execute with:
  - Path: ./job-queue/feature-auth/tasks.md
  - PM_DB_PHASE_RUN_ID: 78
  - Resume from: Task 4

Key behaviors:

  • •Reuses existing phase_run_id (doesn't create new one)
  • •Skips completed tasks (doesn't re-execute)
  • •Continues quality gates (same standards)
  • •Maintains git history (continues commit sequence)
  • •Updates same PM-DB records (complete traceability)

Step 5: Execution Continues

code
Executing remaining tasks...

[4/7] ████████████████░░░░ Task 4 complete ✅
[5/7] ████████████████████ Task 5 complete ✅
[6/7] ████████████████████ Task 6 complete ✅
[7/7] ████████████████████ Task 7 complete ✅

✅ Phase run completed (ID: 78)

📊 Updated Phase Metrics:
   Total runs: 1
   Duration: 72.4 minutes (resumed after 27.2 min)
   Tasks: 7/7 complete

PM-DB Integration

Existing Phase Run

If phase_run exists, resume uses:

python
# Get existing phase_run
run = db.get_phase_run(phase_run_id)

# Don't create new phase_run!
# Reuse existing: PM_DB_PHASE_RUN_ID = run['id']

# For each remaining task:
task_run_id = db.create_task_run(
    phase_run_id=run['id'],  # Same phase_run!
    task_id=task['id'],
    assigned_agent=agent
)

# On completion:
db.complete_phase_run(run['id'], exit_code, summary)

No Phase Run

If no phase_run, create one:

python
# Import to PM-DB first
/pm-db import

# Then start new phase_run
hook_output=$(cat <<EOF | python3 ~/.claude/hooks/pm-db/on-phase-run-start.py
{
  "phase_name": "feature-auth",
  "project_name": "my-project",
  "assigned_agent": "start-phase-execute"
}
EOF
)

phase_run_id=$(echo "$hook_output" | jq -r '.phase_run_id')

Progress Detection Logic

python
def detect_progress(phase_run_id):
    """Detect which tasks are complete"""

    # Get all task_runs for this phase_run
    task_runs = db.list_task_runs(phase_run_id)

    completed_tasks = [
        tr for tr in task_runs
        if tr['exit_code'] == 0 and tr['completed_at'] is not None
    ]

    # Get all tasks from plan
    run = db.get_phase_run(phase_run_id)
    all_tasks = db.list_tasks(run['plan_id'])

    # Find first incomplete task
    completed_keys = {tr['task_key'] for tr in completed_tasks}

    for task in all_tasks:
        if task['task_key'] not in completed_keys:
            return {
                'next_task': task,
                'completed_count': len(completed_keys),
                'total_count': len(all_tasks),
                'completed_tasks': completed_keys
            }

    return {'next_task': None}  # All complete

Use Cases

Scenario 1: Session Dropped Mid-Execution

bash
# Session 1 (interrupted at Task 4):
/feature-new "add auth"
# ... completes Tasks 1-3, then session drops

# Session 2 (resume):
/feature-continue ./job-queue/feature-auth/tasks.md
# ✅ Resumes from Task 4, PM-DB tracking intact

Scenario 2: Intentional Pause

bash
# Day 1: Start feature
/feature-new "payment processing"
# ... completes Tasks 1-5

# Day 2: Resume
/feature-continue ./job-queue/feature-payment/tasks.md
# ✅ Continues from Task 6

Scenario 3: Multiple Attempts

bash
# Attempt 1: Failed at Task 3 (quality gate failure)
/feature-new "admin dashboard"
# ... Tasks 1-2 pass, Task 3 fails quality gate

# Attempt 2: Fix and resume
# (manually fix issues)
/feature-continue ./job-queue/feature-admin/tasks.md
# ✅ Re-runs Task 3 (failed), then continues to Task 4-7

Error Handling

No task-list.md Found

code
❌ Error: task-list.md not found

Path: ./job-queue/feature-auth/tasks.md

Options:
  1. Check path is correct
  2. Run /feature-new to create feature
  3. Create task-list.md manually

No Planning Folder

code
⚠️ Warning: No planning folder found

Expected: ./job-queue/feature-auth/planning/

This suggests the feature hasn't been executed yet.

Options:
  1. Run /start-phase plan first
  2. Run /feature-new for complete workflow
  3. Run /start-phase execute to start fresh

Completed Phase

code
✅ Phase already complete

Phase Run ID: 78
Completed: 2026-01-30 16:45:32
All tasks: 7/7 complete ✅

Nothing to resume.

View results:
  - Dashboard: /pm-db dashboard
  - Summary: ./job-queue/feature-auth/planning/phase-structure/phase-summary.md

Implementation Details

This skill:

  1. •Reads task-list.md to derive paths
  2. •Queries PM-DB for existing phase_run
  3. •Checks task_runs table for completed tasks
  4. •Passes resume context to /start-phase execute:
    • •PM_DB_PHASE_RUN_ID (existing)
    • •Skip tasks (list of completed task_keys)
    • •Resume from (next task_key)

The /start-phase execute skill handles:

  • •Skipping completed tasks
  • •Reusing phase_run_id
  • •Creating task_runs for remaining tasks
  • •Completing the phase_run

Benefits

✅ Session resilience: Resume after crashes/disconnects ✅ PM-DB continuity: Single phase_run tracks complete lifecycle ✅ No duplicate work: Skips completed tasks ✅ Git history preservation: Continues commit sequence ✅ Metrics accuracy: Duration includes all attempts ✅ Quality gate consistency: Same standards throughout

Limitations

  • •Requires PM-DB import before first use
  • •Only resumes from last completed task (not mid-task)
  • •Doesn't handle merged branches (may re-run tasks)
  • •Assumes task-list.md hasn't changed significantly