Problem
Today there's no automation that announces "a new model just scored on ClawBench." Adopters who run the benchmark, submit a PR, and get merged onto the leaderboard get zero downstream marketing for their effort. This is friction we control and could remove.
Proposal
A tiny X/Twitter bot (@clawbench_live or similar) wired to:
- Trigger: GitHub Action on merge to
leaderboard/results.csv (or HF dataset webhook)
- Compose: One-line tweet with the new model's rank + score + trace link
🦞 New on ClawBench V2:
<model> — <intercepted>% Intercepted · <reward>% Reward · rank #<n>/<total>
Trace: claw-bench.com/trace/<id>
- Quote-tweet the adopter's most recent model-release tweet if findable (boosts reciprocal engagement)
- Tag the model provider's official account when known
Implementation
- GitHub Action with
tweepy or twikit — credentials in GitHub Secrets
- Same workflow can mirror to Mastodon + LinkedIn Company if accounts exist
- Optional: also DM the maintainer for moderation pre-post (review window)
Acceptance
Why now
Every model adopter is a free publicity event — but only if the announcement lands somewhere visible. This automates that loop and turns each new submission into a recurring distribution moment.
Problem
Today there's no automation that announces "a new model just scored on ClawBench." Adopters who run the benchmark, submit a PR, and get merged onto the leaderboard get zero downstream marketing for their effort. This is friction we control and could remove.
Proposal
A tiny X/Twitter bot (
@clawbench_liveor similar) wired to:leaderboard/results.csv(or HF dataset webhook)Implementation
tweepyortwikit— credentials in GitHub SecretsAcceptance
Why now
Every model adopter is a free publicity event — but only if the announcement lands somewhere visible. This automates that loop and turns each new submission into a recurring distribution moment.