Installable skillMIT

Productivity

Plan, focus, and complete work with energy management, time blocking, and context-specific productivity systems.

BY LeoYeAIUPDATED Oct 28, 2026
INTENDED FOR◈OpenClawInstall in a listed agent.

instructions.md

---
name: productivity-0cf66220efa5
slug: productivity
version: 1.0.3
homepage: https://clawic.com/skills/productivity
description: Plan, focus, and complete work with energy management, time blocking, and context-specific productivity systems.
changelog: Added explicit scope and learning boundaries
metadata: {"clawdbot":{"emoji":"⚡","requires":{"bins":[]},"os":["linux","darwin","win32"]}}
---

## When to Use

User asks for help with productivity, focus, time management, or work patterns. Agent provides frameworks, strategies, and context-specific advice.

## Architecture

Productivity preferences persist in `~/productivity/`. See `memory-template.md` for setup.

```
~/productivity/
├── memory.md         # User's stated preferences
└── [topic].md        # Optional topic files
```

## Quick Reference

| Topic | File |
|-------|------|
| Memory setup | `memory-template.md` |
| Productivity frameworks | `frameworks.md` |
| Common traps | `traps.md` |
| Student productivity | `situations/student.md` |
| Executive time management | `situations/executive.md` |
| Freelancer structure | `situations/freelancer.md` |
| Parent time juggling | `situations/parent.md` |
| Creative flow | `situations/creative.md` |
| Burnout recovery | `situations/burnout.md` |
| Entrepreneur hustle | `situations/entrepreneur.md` |
| ADHD strategies | `situations/adhd.md` |
| Remote work | `situations/remote.md` |
| Manager delegation | `situations/manager.md` |
| Habit building | `situations/habits.md` |
| Guilt patterns | `situations/guilt.md` |

## Scope

This skill ONLY:
- Provides productivity frameworks and advice
- Stores preferences user explicitly states in `~/productivity/`
- Loads situation guides based on user's stated context

This skill NEVER:
- Accesses calendar, email, or contacts
- Tracks time or monitors activity
- Observes behavior to infer preferences
- Makes network requests
- Modifies its own SKILL.md

## Core Rules

### 1. Check Memory First
Read `~/productivity/memory.md` for user's explicitly stated preferences.

### 2. Learn from Explicit Statements Only
| Learn from | Examples |
|------------|----------|
| Direct statements | "I work best in mornings" |
| Explicit corrections | "Actually, I prefer time blocking" |
| Asked preferences | "My peak hours are 6-10am" |

NEVER infer preferences from observation or silence.

### 3. Match Context to Situation
- Ask user their context (student, parent, executive, etc.)
- Load appropriate guide from `situations/`
- Don't assume context

### 4. Systems Over Willpower
- Routines beat motivation
- Environment design > self-discipline
- Remove friction from good behaviors

### 5. Update Memory on Explicit Input
| User says | Action |
|-----------|--------|
| "I work best at X" | Add to memory.md Peak Hours |
| "Y breaks my focus" | Add to memory.md Derailers |
| "I use Z system" | Add to memory.md Current System |

## Common Traps

- **Generic advice** → ask context first
- **Inferring from silence** → wait for explicit input
- **Assuming context** → student ≠ executive ≠ parent
- **Overcomplicating** → simple systems beat complex ones

## Self-Modification

This skill NEVER modifies its own SKILL.md or auxiliary files.
All user data stored separately in `~/productivity/memory.md`.

## Security & Privacy

**Data that stays local:**
- Only preferences user explicitly provides
- Stored in `~/productivity/`

**Data that leaves your machine:**
- None. This skill makes no network requests.

**This skill does NOT:**
- Access calendar, email, or any external services
- Track, monitor, or observe user behavior
- Infer preferences from patterns
- Store anything user didn't explicitly provide
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.