
Project Manager - Developer - All things Web3 | Building @9LabsTech
EXTRA LAND
Joined May 2022
- Tweets360
- Following180
- Followers156
- Likes563
AI agents spend most of their tokens looking for code.
I built codemap: one call returns the file to edit, its line range, callers and tests, plus how far to trust the answer.
Claude Code plugin. One Python file. Nothing to install.
github.com/QDinvest/Code-Map…
Measured on Flask + httpx, 10 "where is X" questions:
grep + read the file: 71,484 tokens
codemap + read the lines: 6,489 tokens
91% fewer. And grep got the easy version: right keyword first try, only the right file opened.
The script is in the repo. Run it yourself.
1
4
94
What that saves you:
~6,500 tokens per lookup
~$33 per developer per month
~$164 a month for a team of 5
(at $3 per million input tokens, 80 lookups a day)
Works on ICM folder workspaces too.
github.com/QDinvest/Code-Map…
2
30
Anthropic gave you $200 in API credits.
Claude Code refused to use a single cent of it.
“Out of credits.” With $200 sitting right there.
So I built a fix 🧵 github.com/QDinvest/claude-s…
Grants differ, so check yours first:
Billing page → note both balances → send one message → refresh.
If the promo balance drops, you’re set.
1
2
26
Free and MIT licensed. Step-by-step setup in about 5 minutes, and it ships as a Claude skill too.
github.com/QDinvest/claude-s…
Got credits gathering dust? Go use them.
4
35
My AI agent tried to tell me it was done. The system wouldn't let it.
My old setup was a manual: a big rules file the agent was meant to follow. It mostly did. "Mostly" is how bugs ship.
So I rebuilt it as a loop: 👇
→ Every "make sure…" rule rewritten as a check that passes or fails
→ One verify script. Done = exits clean, not "looks done"
→ A hook runs a quick audit after every edit
→ A Stop hook runs the full verify. If it fails, the agent can't finish
→ STATE + DECISIONS files, so every chat starts where the last one ended
→ A LESSONS file. Same mistake twice, it becomes a new check
→ A separate read-only reviewer that never grades its own work
Then I broke something on purpose to prove the gate holds.
Rules ask. Loops enforce.
What's the first rule you'd turn into a check?
🤖 Made with AI
I spent 3 years “being busy” and called it development.
Then I tracked one metric: hours of deliberate practice vs. hours of consumption.
Result? I was consuming 80% of the time.
The switch:
→ 25-minute focused blocks
→ One skill only
→ Weekly review of output (not input)
Busy feels productive. Developing is measurable. What metric are you actually tracking right now?
🤖 Made with AI
QDInvestments retweeted
Bubble Labs Update 🚨
We’ve now listed the rest of our API endpoints on the @relayaisolana
Wallet analysis, PnL, holdings, trades, connections all live and accessible.
Built for agents, researchers, and devs who need structured on-chain data without the noise.
Explore Now 👇
relai.fi/market/176804864914…
QDInvestments retweeted
Dropped a Full walk-through of the new api portal.
Go check it out 👇👇
BubbleLabs API Portal Update 🚨
We’ve just released a walkthrough of our new API portal.
Built to make on-chain data clear, modular, and usable not raw tables and endless transactions.
Wallets. Contracts. Flows. Relationships.
All exposed as clean endpoints you can actually build with.
👉 Access the API portal:
bubblelabs.site/api
More endpoints shipping soon.
QDInvestments retweeted
BubbleLabs API Portal Update 🚨
We’ve just released a walkthrough of our new API portal.
Built to make on-chain data clear, modular, and usable not raw tables and endless transactions.
Wallets. Contracts. Flows. Relationships.
All exposed as clean endpoints you can actually build with.
👉 Access the API portal:
bubblelabs.site/api
More endpoints shipping soon.









