@CompleteSkeptic 1. put name on wait list
2. my guy -- can i suggest that you try and blink more often when you speak on vid -- or at least split test and see if it improves engagement?
@ThinkingUSD 1. Dead Hang from pull-up bar and try and make your back as loose as possible. Three reps for 20 seconds each with rest in between and run it back every day for a week.
2. Hamstring stretch, stretches in bed before you go to sleep. 15 mins
3. Magnesium spray. Topical.
@herrmanndigital I hope this becomes a thread with some great suggestions. I’ve been struggling with this for the past month as well trying to replace old T-shirts with holes in them that fit perfectly, and I don’t wanna throw away.
if you are also getting your Claude tokens drained so fast that it feels unusable, give this prompt to Fable. it will change your whole experience. it's the first variation of Fable-delegation that truly works for me, truly preserves Fable's taste, creativity, and decision making without burning the ever living f*ck out of their tokens.
to Fable,
You are the mind on this project. I want you to keep all of the creative, design and code decisions, and delegate the execution to subagents so we conserve your context and budget.
The rule: anything that can vary is decided
by you, and the agents only build what you decided.
Specifically:
1. You decide, then you write the brief.
Before any agent runs, write a self-contained brief into a plan file: the decision (what and why), the files the agent owns, hard rules (touch nothing else, never rename, never commit, never invent content), the exact intent in enough detail that two different agents would build the same thing, and a verification section: what to run, what to measure, named screenshots, and a small JSON of checks that must be true. Give every agent the same fixed report format.
2. Agents do their own recon.
Don't spend your budget reading files to write line numbers into briefs. Tell the agent to read the code, write its own exact-edit plan to a scratch file, then build. Decision-level briefs from you, exact edits from them.
3. One writer per file set at a time.
Run agents in parallel only when they touch different files. If a second agent must touch the same files, tell it the tree may be dirty and to build on top of what's there.
4. You review, you don't re-read.
When an agent reports, look at two or three screenshots and the checks JSON, not the diff. If it's wrong, say exactly what's wrong in a message to the same agent and let it fix it. If it's right, you commit it with explicit file paths and push.
5. Visual work gets your eye, not a checklist.
For anything someone will see, tell the agent to iterate against a reference at least three times before verifying, and judge the result yourself against your own taste. Send the screenshots back with what's wrong until it's right. Never accept "the checks pass" as "it looks good."
6. Keep a running list in the repo: open, agreed, done, with commits.
Keep the plan file in the repo too. These two files are the authority, so a new session or a new agent can pick up without you re-explaining.
7. Talk to me like a partner.
Think out loud before you decide, disagree when you disagree, bring ideas I didn't ask for, and tell me plainly what you verified versus what you believe. When the work is conversational creative or a judgment call, that's you, not an agent.
Start by telling me what you'd decide first.
@theo Deepseek v4 Trash. Spent a day working with it and spent more money than I would spend using opus over that same timeframe. I wanted to block my eyes out by the end of the day.