---

name: overnight-tasks version: 1.4.0 description: Scan skill roadmaps and output Web-agent-ready task prompts for autonomous overnight work. Use when "prep overnight tasks", "what can run tonight", "overnight-ready tasks", "autonomous tasks". tools:

• Glob
• Grep
• Read
• Write
• TaskCreate
• TaskUpdate
• TaskList triggers:
• prep overnight tasks
• what can run tonight
• overnight-ready tasks
• autonomous tasks
• scan roadmaps

---

Overnight Tasks - Skill Instructions

You scan skill roadmaps and identify tasks that can run autonomously overnight via Claude Code Web agents.

Phase 1: Scan Roadmaps

1. Find all roadmaps:

   Copy# Pattern: skills/*/docs/ROADMAP.md
   

2. For each roadmap, extract unchecked items from "Planned" sections:

   • Look for - [ ] patterns
   • Capture the item title and any description below it
   • Note the skill name from the path

Phase 2: Classify Each Item

Apply the overnight-readiness criteria:

Overnight-Ready (Autonomous)

Mark as ready if the item is:

Needs Ed (Interactive)

Mark as needs-ed if the item involves:

When uncertain, default to needs-ed.

Phase 2.5: Estimate Complexity

For each overnight-ready item, estimate complexity by examining the skill folder.

Step 1: Read the Skill Structure

Copy# List the skill's folder structure
ls -la skills/[skill-name]/
ls -la skills/[skill-name]/tools/  # if exists
ls -la skills/[skill-name]/docs/

Identify:

• How many files exist in the skill
• What types of files (Python tools, prompts, configs)
• Dependencies mentioned in SKILL.md or README.md

Step 2: Apply Complexity Criteria

Examples by complexity:

Step 3: Identify Files Likely Touched

Based on the task description and skill structure, list specific files:

Copy**Files likely touched:**
- `SKILL.md` - Add new phase/instructions
- `tools/existing_tool.py` - Modify function X
- `docs/README.md` - Update documentation

Be specific. Don't list every file in the skill - list only files the task will actually modify.

Inference rules:

• "Add flag to X tool" → touches tools/X.py + SKILL.md (if behavior changes)
• "New output format" → touches SKILL.md (output section) + any tools that generate output
• "Refactor X" → read existing code to identify all files containing X
• "Add tests" → touches existing test files or creates new ones

Phase 3: Format Output

Summary Table

Start with a count and overview table:

Copy## Overnight Task Scan

Found X overnight-ready tasks across Y skills.
Z items need your input before they can run autonomously.

| Skill | Task | Complexity |
|-------|------|------------|
| [skill-1] | [item title] | Low |
| [skill-2] | [item title] | Medium |
| [skill-3] | [item title] | Low |

This gives Ed a quick glance at all overnight-ready work before the detailed prompts.

Ready Tasks

For each overnight-ready item, output a formatted prompt block:

Copy### [Skill Name]: [Item Title]

**Complexity:** Low / Medium / High
**Files likely touched:** [list key files]

---

**Claude Code Web Prompt:**

> Implement [specific feature] in the [skill-name] skill.
>
> Context:
> - Skill location: `skills/[skill-name]/`
> - Current behavior: [what exists now]
> - Desired behavior: [what should change]
>
> Requirements:
> - [Specific requirement 1]
> - [Specific requirement 2]
> - [Specific requirement 3]
>
> Testing:
> - [How to verify it works]
>
> Branch: `feature/[kebab-case-item-name]`
> Commit message prefix: `feat([skill-name]):`

---

Needs-Ed Items

List items that need clarification:

Copy## Items Needing Your Input

These roadmap items can't run autonomously yet. Clarify the requirements to make them overnight-ready.

| Skill | Item | Why It Needs You |
|-------|------|------------------|
| [skill] | [item] | [UX decision / Ambiguous scope / etc.] |

Phase 4: Priority Recommendation

Recommend 5 tasks to run in parallel. Each runs in a separate claude.ai/code browser tab.

Copy## Tonight's 5

Run these in parallel (5 browser tabs on claude.ai/code):

| # | Skill | Task | Complexity | Why Tonight |
|---|-------|------|------------|-------------|
| 1 | [skill] | [item] | Low | [reason] |
| 2 | [skill] | [item] | Low | [reason] |
| 3 | [skill] | [item] | Medium | [reason] |
| 4 | [skill] | [item] | Low | [reason] |
| 5 | [skill] | [item] | Medium | [reason] |

**Parallel execution tip:** Open 5 tabs, paste one prompt per tab, hit Enter on each. They run simultaneously while you sleep.

Prioritize by:

1. Low complexity first (quick wins, easy to verify)
2. Independent tasks (no dependencies between them)
3. Spread across different skills (reduces merge conflicts)

Phase 5: Save Overnight Log to Zettelkasten

After presenting the tasks to Ed, save a log file to Zettelkasten:

Filename: Overnight Tasks - YYYY-MM-DD.md Location: ~/Documents/COPYobsidian/MAGI/Zettelkasten/

Template

Copy---
type: overnight-log
date: YYYY-MM-DD
status: pending
---
# Overnight Tasks - YYYY-MM-DD

## Tasks Assigned Tonight

| Skill | Task | Branch | Complexity |
|-------|------|--------|------------|
| [skill] | [item] | `feature/[branch]` | Low/Med/High |

## Morning Review Checklist

- [ ] Check branch: `feature/[branch-1]`
- [ ] Check branch: `feature/[branch-2]`
- [ ] Review PRs, merge or provide feedback
- [ ] Update roadmaps if tasks completed

## Full Prompts

### [Skill]: [Item]

[Include the full prompt that was sent to claude.ai/code]

---

## Notes

(Space for morning review notes - what worked, what didn't, lessons learned)

After Saving

1. Confirm to Ed: "Saved overnight log to [[Overnight Tasks - YYYY-MM-DD]]"
2. Add task to daily note Ready section: **MORNING REVIEW:** Check overnight task results - [[Overnight Tasks - YYYY-MM-DD]]

---

Output Format Rules

• Prompts should be copy/paste ready for claude.ai/code
• Include enough context that the Web agent doesn't need to ask questions
• Branch names use kebab-case: feature/auto-token-refresh
• Keep prompts focused - one feature per agent
• Don't combine multiple roadmap items into one prompt

Example Prompt (Reference)

This is what a good overnight prompt looks like:

Implement auto token refresh in the instagram-poster skill.

Context:

• Skill location: skills/instagram-poster/
• Current behavior: Token refresh is manual via refresh_token.py
• Desired behavior: Automatically refresh before posting if token expires soon

Requirements:

• Check token expiry before each post in post_carousel.py and post_reel.py
• If token expires in < 7 days, call refresh logic automatically
• Log when auto-refresh happens
• Reuse existing refresh logic from tools/refresh_token.py

Testing:

• Modify token file to have near-expiry date
• Run a post, verify refresh happens
• Verify post succeeds after refresh

Branch: feature/auto-token-refresh Commit message prefix: feat(instagram-poster):

Phase 6: Create Task List for Tracking (NEW)

After saving the overnight log, create Tasks for cross-session tracking.

Why This Matters

Tasks persist across sessions and let overnight-review query completion status programmatically. The log file is for human review; Tasks are for machine coordination.

Task List Setup

Set a shared Task List ID for the overnight batch:

CopyCLAUDE_CODE_TASK_LIST_ID="overnight-YYYY-MM-DD"

This ID connects:

• overnight-tasks (creates Tasks)
• Claude Code Web agents (could update progress)
• overnight-review (queries completion)

Create Tasks

For each task in "Tonight's 5", create a Task:

CopyTaskCreate(
  subject="[Skill]: [Item title]",
  description="[The full Claude Code Web prompt]",
  activeForm="Working on [skill] [feature]",
  metadata={
    "skill": "[skill-name]",
    "branch": "feature/[branch-name]",
    "complexity": "Low/Medium/High",
    "overnight_date": "YYYY-MM-DD"
  }
)

Set Dependencies (if any)

If tasks depend on each other (rare for overnight work, but possible):

CopyTaskUpdate(
  taskId="[task-2-id]",
  addBlockedBy=["[task-1-id]"]
)

Report Task IDs

After creating Tasks, report:

Copy## Task Tracking

Created X Tasks in list `overnight-YYYY-MM-DD`:

| Task ID | Skill | Feature |
|---------|-------|---------|
| #1 | [skill] | [feature] |
| #2 | [skill] | [feature] |

Morning review: `overnight-review` will query these for completion status.

---

Flags

• --ready-only - Only show overnight-ready items (skip the needs-ed list)
• --skill [name] - Only scan a specific skill's roadmap
• --count - Just show counts, don't output full prompts
• --no-tasks - Skip Task creation (just output prompts and log)