i let grok bot run my internship and full-time applications.
10 a day, from a routine i set up once.
anyone else automating the job hunt, or am i the only one?
send this prompt to claude code when a repo starts feeling messy. it should audit the project and update your claude md file too, not only the code.
read the repo and your claude md file.
1. list what's broken or stale in two buckets: sure fixes, and things i should review.
2. only patch the sure fixes. keep each patch small and in the existing style.
3. if a sure fix changes how we should work in this repo, update the claude md file to match. do not invent new rules.
4. for the review bucket, write a short list with file paths and why. do not touch that code.
5. show me both buckets and wait before you start patching.
how to make youtube shorts for your app without opening a video editor
the play: let two agents pass the work back and forth. grok bot writes and schedules, claude code refines and renders.
my pipeline for linya:
1. grok bot writes the basic scripts.
2. claude code refines them with my viral carousel playbook skill.
3. claude code makes the shorts with the shorts-animator skill.
https://t.co/nCDPPzdavA
4. claude code spits out the file path where each reel is saved on my laptop.
5. grok bot reviews them and writes a prompt for any changes.
6. grok bot schedules them automatically through the buffer mcp.
the catch: i still copy and paste prompts between grok bot and claude code by hand.
@loganstakeley if im not wrong, u can only generate 'unlimited' videos in the higgsfield web app, generating them in claude via the mcp will charge u credits
what's one claude code skill you actually reuse, not just a prompt you tried once and forgot?
drop the name or what it does. looking for ones that survived more than a single session.
every time you start a feature in claude code, do this first:
write a 5-line brief yourself. the change, the files that can move, the files that cannot, how you'll know it's done, and what the agent should do if it's unsure.
paste that at the top of the session. if you let the model invent the plan, you've already given away the part you were supposed to decide.
one tip for claude code: put a fence in the prompt, not just the goal.
bad: "refactor the payments module"
good: "refactor the payments module. do not change the public api. do not add a new dependency. show me a 5-bullet plan before you edit anything."
if the prompt has no hard no's, the model will invent some scope for you.
if you run a "find everything broken" sweep on a repo, should it also update your claude md file and the docs, or only the code?
curious where people draw that line.
vibe coding isn't the problem. vibe coding without tests is the problem
the play: let ai write the code, but never let it decide when it's done
> before the prompt
- write down what "working" means in 3 to 5 plain lines
- turn those lines into tests first, or have claude code write only the tests from them
- read the tests yourself. if you can't follow them, neither can the agent
> while it builds
- put "run the tests before you say done" in your CLAUDE.md
- paste the failing test output back, not "it's broken"
- one feature per session so a red test points at one thing
> before you ship
- run the full suite, not just the new tests
- open the app and click through the main flow once yourself
> the catch
- tests the agent writes for its own code barely count. that's grading your own homework
- write them before the code, or in a fresh session that never saw the code
ai made writing code cheap. checking it is the job now
bookmark this before your next vibe coding session
there is a skill called vox-explainer
it builds a vox-style documentary reel in code, then exports it to mp4. no video generator, no stock footage.
i ran it on a fitting-room story for my app. this is what it made.
https://t.co/Rf9HRNEMxA