Holy shit, the most-starred Claude skill repo on GitHub is not a skill pack, it is a memory that claims to double your plan.
Open GitHub's claude-skills topic. Number one is claude-mem with 92.5k stars. It keeps context across sessions, so the agent remembers what it did yesterday.
Now the money. Its README claims ~10x token savings and "up to 100% more usage from your plan". The code is free under Apache-2.0, but the default hosted memory is a 30-day trial, and the price after it is not on the page.
The rest of the top 20 says the same thing:
planning-with-files, 26.4k, MIT. Plans live in files, and one release cut per-session overhead from about 2,254 to 1,042 tokens.
memU, 14.4k. Memory across agents.
alirezarezvani/claude-skills, hundreds of skills, only number 11 at 25.1k.
The savings claims come with no method. claude-mem's page also lists a third-party crypto token the creator "officially embraced". Stars are bookmarks, not proof.
The cheapest token is the one your agent does not spend re-learning yesterday.
@Grirku Simple rule: value per dollar for my actual daily work, not hype.
Claude Pro at $20 with Claude Code included does most of my coding — easy S. Ultra at $200 that I barely touch ends up in F. Free tiers that keep cutting me off mid-task land low too
I asked Gemini and Claude to make a 15-second motion designer showreel.
Gemini signed it "GEMINI".
Claude signed it "CLAUDE".
Ego is the one skill they both nailed.