Buddy on Gemini CLI and Claude Code

Source: YouTube auto-generated captions, lightly cleaned (repeated caption lines removed, misheard brand names corrected).

0:00 — Half the budget wasted

Most accounts leak money on search terms that never convert. This morning, I pulled 30 days from one live account under management. Total spend was $1,366. $634 of that went to search terms with zero conversions. That is nearly half the budget buying nothing. Average CPA sat at $59 on the terms that did convert. Those zero conversion queries are not a mystery list. The same poll surfaced 85 distinct terms that are negative keyword candidates right now.

Leaving them active means tomorrow's budget repeats today's waste. A media buyer who waits for the next weekly audit keeps paying for the same dead traffic while CPA stays stuck near $59.

0:47 — Only plugin on both

The practical fix used to live only inside custom scripts and agency desks. Buddy is now listed on the Gemini CLI marketplace and the Claude code marketplace. It is the only Google Ads plugin on both one advertising agent reachable from the two terminals serious operators already keep open during build and review work. Step one is accepting

1:08 — Step 1 — Install one agent

that you install one agent not two parallel bots. Buddy is the same product on both marketplaces. So skills, memory patterns and Google ads or stay consistent. You avoid teaching one workflow in Gemini CLI and a different one in clawed code. That single identity is why later steps do not fork into conflicting playbooks. Operators split attention when each CLI owns a different ads helper. Shared identity means a negative keyword path started in one terminal can be reviewed in the other without reexplaining account structure.

You also keep one audit trail for what changed in Google Ads. That is the reason the install order starts with the agent, not with terminal preference. Step two is opening the two terminals

1:55 — Step 2 — Open two terminals

you already use for code and analysis. Gemini CLI and Claude code are not competing installs here. They are two doors into the same buddy agent. Desk work, long refactors, and quick checks can sit in different shells while the Google Ads context remains one source of truth instead of two half-sync helpers. Use whichever terminal matches the task in front of you. Long structured cleanup can live in one shell while a fast spend check runs in the other. Because Buddy is the only Google Ads plugin on both marketplaces, you are not translating prompts between vendors. The why is simple. Less context switching tax when the account is already leaking $634 a month. Step three

2:41 — Step 3 — Load 77 skills

is loading the skill set, not inventing prompts from scratch. Buddy ships 77 skills aimed at real Google ads work. search terms, negatives, RSA checks, budget reads, and account hygiene. Skills matter because a bare chat window forgets process under pressure. A named skill keeps the agent on the audit path that found those 85 negative candidates. Think of skills as short run books the agent already knows how to execute. You call the skill that matches the leak, then read the output against live account numbers. That beats pasting a giant prompt and hoping the model remembers your negative keyword rules.

The order puts skills after terminals. So, install and orth are done before you trust any recommendation. Step four is asking buddy against the live account,

3:29 — Step 4 — Ask the live account

not a demo file. Point it at the same 30-day window that showed $1,366 spend and $634 wasted on zero conversion terms. The agent should surface the 85 negative candidates with enough query text to act. Live context is the point of an ads plug-in. Canned screenshots cannot reorder your budget. When the readout lands, anchor on the same four numbers every time. Spend, wasted spend, negative candidate count, and average CPA. This morning that was $1,366, $634, $85 to $59 CPA. If a skill returns narrative without those anchors, run it again. Media buying decisions need magnitudes, not adjectives before anyone touches the negative list. This order matters because each stage removes a

4:26 — Why this order matters

failure mode. One agent prevents split brains. Two terminals meet you where work already happens. 77 skills stop prompt drift live asks bind advice to real spend. Skip ahead to a clever question with no orth and no skill and you get generic copy while $634 keeps burning. Sequence is how the waste actually shrinks. Step five is shipping the negatives, not filing the report.

4:52 — Step 5 — Ship the negatives

Take the 85 candidates, confirm match types, and add them so the next 30 days cannot repurchase the same zero conversion traffic. Buddy shortens the fine step. The buyer still owns the apply step. That handoff is deliberate. Agents surface waste quickly. Account permissions and policy judgments stay with the human. After the list is live, repull a short window instead of waiting a full month. You want early proof that wasted search spend is falling without breaking the queries that still hold the $59 CPA together. A 7-day check is enough to see direction. If waste is flat, reopen the skill and widen the candidate logic rather than guessing in the UI by hand. Make the pair of terminals part of the weekly habit, not

5:40 — Keep the weekly habit

a one-off stunt. Monday waste pass in one shell, midweek RSA or budget pass in the other, same buddy agent both times. The only Google Ads plugin present on both marketplaces is what makes that habit cheap to keep. Tools you open twice a week outperform tools you admire once. Do not stand up a second ad spot beside buddy just because a terminal feels empty. Do not paste the 77 skill names into a free form chat and call that an install. Do not accept a readout that skips spend, waste, candidate count, and CPA. Those shortcuts recreate the same blind spot that left 85 negatives sitting idle while $634 left the account. Recap in one line of practice.

One buddy agent on Gemini CLI and Claude

6:30 — Recap

Code, the only Google Ads plug-in on both marketplaces. 77 skills, then a live ask on the account that spent $1,366 and wasted $634 across 85 dead terms at a $59 cost per acquisition. Install, open both doors, load skills, ask live, ship negatives. That is the whole loop. If you want the written walkthrough with marketplace links and the skill framing, the blog is on ahmeego.com. Use the live numbers from your own last 30 days, not hours. Start with waste and negatives before creative experiments.

Body by Army Go, a company of It All Started with an idea.