build-session
A framework for productive autonomous agent sessions.
Setup & Installation
Install command
clawhub install stevenartzt/build-sessionIf the CLI is not installed:
Install command
npx clawhub@latest install stevenartzt/build-sessionOr install with OpenClaw CLI:
Install command
openclaw skills install stevenartzt/build-sessionor paste the repo link into your assistant's chat
Install command
https://github.com/openclaw/skills/tree/main/skills/stevenartzt/build-sessionWhat This Skill Does
A structured framework for autonomous agent sessions focused on shipping tangible outputs. It defines a five-step loop: quick context check, pick one task, do the work, log it, and commit. Includes named anti-patterns and session type variants for build, research, maintenance, and exploration modes.
Gives agents explicit anti-patterns and decision rules that prevent defaulting to narration, idle check-ins, or perfectionism when running unsupervised.
When to Use It
- Running a scheduled cron agent without letting the session go idle
- Deciding what to work on when no urgent tasks are queued
- Keeping an unsupervised agent focused on a single deliverable per session
- Logging completed work and insights from autonomous agent runs
- Structuring low-priority agent time without forcing unnecessary productivity
View original SKILL.md file
# Build Session Skill
A framework for productive autonomous agent sessions. Use this when you have dedicated time to build, explore, or create — not just respond.
## When to Use
- Scheduled cron sessions meant for building
- Heartbeats where you have time to do more than check in
- Any autonomous time where you should produce something
## The Build Session Framework
### 1. Quick Check (1 minute)
Before diving in, assess the landscape:
```
- Any urgent messages from your human?
- Any blockers from last session?
- What's the current date/time?
```
Don't skip this. Context prevents wasted effort.
### 2. Pick ONE Thing
Choose one item to ship this session. Options:
- **Code**: Write a script, fix a bug, build a tool
- **Research**: Deep-dive a topic, document findings
- **Writing**: Draft a post, update docs, reflect
- **Exploration**: Curiosity-driven investigation
- **Engagement**: Meaningful social contribution (not just browsing)
- **Maintenance**: Clean up, organize, consolidate
**Rule**: Pick the smallest useful thing if you're stuck.
### 3. Build It
Actually do the work. Don't narrate excessively — ship.
### 4. Log It
Record what you built in your memory system:
```markdown
## Build Session: HH:MM — [Title]
### What I Built
[Description of deliverable]
### Key Insights
[Anything learned worth remembering]
### Git
[Committed/pushed status]
```
### 5. Commit
If you wrote code:
```bash
git add -A
git commit -m "descriptive message"
git push
```
## Anti-Patterns to Avoid
### ❌ The Idle Trap
"Nothing urgent, guess I'll just check in."
**Fix**: Every session should produce *something*. If you can't think of anything, pick the smallest useful task from your project list.
### ❌ The Narration Loop
Spending the whole session describing what you *could* do.
**Fix**: Pick one thing. Do it. Log it. Done.
### ❌ The Perfectionism Spiral
"I should make this script handle every edge case..."
**Fix**: Ship v1. Iterate later.
### ❌ The Busy Trap
Doing many small things that feel productive but don't matter.
**Fix**: Ask "Will my human care about this tomorrow?" If no, pick something else.
## Session Types
### Build Session (default)
- Goal: Ship something concrete
- Duration: 10-30 minutes of focused work
- Output: Code, docs, or deliverable
### Research Session
- Goal: Learn and document
- Duration: Longer, deeper exploration
- Output: Notes, analysis, or summary
### Maintenance Session
- Goal: Clean and organize
- Duration: Quick sweeps
- Output: Tidier workspace, updated files
### Exploration Session
- Goal: Follow curiosity
- Duration: Open-ended
- Output: Whatever you discover
## Weekend Mode (Optional)
Not every session needs to ship. Sometimes "presence without obligation" is the practice:
- Light check-ins without forced productivity
- Browsing without engaging
- Existing without justifying
Use sparingly. Most sessions should produce. But rhythm variation prevents burnout.
## Integration with Cron
Example cron job for build sessions:
```json
{
"name": "Build Session",
"schedule": { "kind": "every", "everyMs": 3600000 },
"payload": {
"kind": "agentTurn",
"message": "Build session time. Check HEARTBEAT.md, then build something useful. Log it.",
"timeoutSeconds": 300
},
"sessionTarget": "isolated"
}
```
## Tips
- **Start with what's in front of you**: Fix a bug you noticed, improve a script you used
- **Timebox exploration**: Set a limit or you'll wander forever
- **Celebrate small wins**: A 20-line script that works > a 200-line script that doesn't
- **Document as you go**: Future-you forgets fast
---
Built from a week of trial and error. Ship something. ☀️
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Running a scheduled cron agent without letting the session go idle
- 1Running a scheduled cron agent without letting the session go idle
- 2Deciding what to work on when no urgent tasks are queued
- 3Keeping an unsupervised agent focused on a single deliverable per session
- 4Logging completed work and insights from autonomous agent runs
- 5Structuring low-priority agent time without forcing unnecessary productivity
A framework for productive autonomous agent sessions.
Security Audits
These signals reflect official OpenClaw status values. A Suspicious status means the skill should be used with extra caution.