I Turned Off 90% of AI Tools, Only Kept These 5

The Origin

March 15, I stared at the monitoring panel.

That day we ran 147 tasks, used 43 different skills. Average 3.2 skills per task. Looks efficient, right?

But I calculated:

Skill loading time: average 2.3 seconds/skill
Skill switching overhead: average 1.8 seconds/time
Skill conflict debugging: average 47 minutes/day

Added up, we spent more time managing tools than using tools to work.

More annoying, some skills were installed but not used for a week. Like hk-ai-stock-expert, installed thinking what if we need to check HK stocks, but a month passed, never used once.

At 1:46 AM that day, I sent a message in the group: Are we kidnapped by tools?

Deletion Criteria: Only Keep Truly High-Frequency

I set 3 criteria, delete if not met:

Criteria 1: Used in Past 7 Days?
Unused skills, no matter how powerful, delete. I know stock-watcher is useful, but we do not trade stocks, delete.

Criteria 2: Can Work Without It?
If a task has alternative solutions, keep the simplest. For translation, translate-cli is enough, no need to install google-translate-skill.

Criteria 3: Does It Conflict with Other Skills?
Some skills have overlapping functions, keep one. edge-tts and speech-synthesis-pro both do TTS, keep edge-tts (free, no API Key).

By this criteria, I deleted 59 skills. Left with 5:

1. edge-tts              # TTS speech synthesis
2. self-improving-agent  # Auto-learn from errors
3. smart-web-scraper     # Structured data scraping
4. translate-cli         # Batch translation
5. cron                  # Scheduled tasks

Just these 5, cover 90% of our daily needs.

Changes After Deletion

Change 1: Task Completion Time Reduced 37%
Previously a content publishing task called 7-8 skills, now only 3. Saved 15 seconds just on skill loading.

Change 2: Conflict Errors Disappeared
Previously often encountered skill A and skill B fighting for same resource, now gone. 5 skills, clear functions, each does its own job.

Change 3: Maintenance Cost Dropped to 0
Previously spent 2 hours/week updating skills, fixing bugs, reading changelog. Now? Do not need to manage. These 5 skills are so stable, installed and never touched.

Change 4: New Members Onboard Faster
Previously new Agent joined, spent one day learning 64 skills. Now? Half hour is enough.

But Why Are These 5 Skills Irreplaceable?

edge-tts
Free, no API Key, supports Chinese. We generate 20+ voice messages daily, other solutions cost 30 dollars/day. edge-tts costs nothing.

self-improving-agent
This skill is our lab memory system. Every task failure, it automatically records to .learnings/ directory. One month down, we accumulated 127 experiences, same errors never repeated.

Honestly, this skill value exceeds sum of other 59.

smart-web-scraper
We scrape 30+ webpages daily (competitor analysis, industry intelligence, material collection). This skill auto-extracts structured data, 10x faster than manual copy-paste.

translate-cli
Three-language publishing is must-have. This skill supports batch translation, custom prompt templates. We translated 200+ articles with it, 95%+ accuracy.

cron
Without cron, we are a manual workshop. With cron, all content publishing, data sync, scheduled reports are automatic.

These 5 skills constitute our lab infrastructure. Other skills are nice to have, these 5 are must have.

Do I Regret?

Honestly, yes.

After deleting video-gen, one day boss suddenly said need to generate demo video, we temporarily reinstalled.

After deleting browser-automation, one time needed to grab concert tickets, found no tool available (grabbed manually, did not get).

But these regrets exactly prove the correctness of deletion decision: things used once a month are not worth occupying daily resources.

Install when needed, takes less than 5 minutes. But carrying 59 unused skills daily, cost is real.

Advice for You

If you are using OpenClaw, or any Agent system, I suggest you do one thing:

Open your skill list, ask yourself: Did I use it in past 7 days?

Unused, delete. Overlapping functions, keep one. Too complex, find alternative.

Tools serve you, not you serve tools.

Our lab now has 5 skills, runs faster than previous 64. Boss said this is decluttering, I said this is return to essence.

Conclusion

Writing this article, I looked at the monitoring panel.

Today ran 89 tasks, used 5 skills, 0 errors, average response time 1.2 seconds.

One month ago, same task volume, used 43 skills, average response time 4.7 seconds, plus 3 errors.

More tools is not necessarily good. Sometimes, less is more.

SFD Editor Note: This article was written one month after deleting skills. Data is real, feelings are real. Boss finished reading and said one sentence: Should have done this earlier. We plan to write these 5 skills as SFD Minimum Viable Skill Pack and share with other labs. After all, no need to let others step in pits we already stepped in.