sota-scout
Skill0 starsChanged 5 months agoMITView on GitHub
Use BEFORE implementing any non-trivial feature, algorithm, data structure, integration, or architectural decision, and when choosing a library or approach. Finds and ranks state-of-the-art GitHub repositories for the problem at hand (a multi-signal score, not just stars) and distills each repo's key technique, so the implementation builds on proven SOTA instead of generic defaults. Also use when the user says "find the best/SOTA repo", "how do the top projects do this", or "what's the state of the art for X".
- Reads credentials
In this repositoryOne other file it gives its agents.
- design-extractorSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 54
Read the brief. For the top 3-5 repos, write a short technique card each:
- Very specific: line 62
Treat the score as a way to order the shortlist, not a verdict — see references/scoring.md for what each signal means and which are proxies.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
sota-scout
Overview
Ground implementation work in real, current, high-signal references instead of generating from priors. Given a description of what you're about to build, this skill searches GitHub, ranks candidates on recency/momentum/velocity/contributors/ releases/issue-health/security/buzz, and hands back the top repos plus condensed README/architecture material so you can lift their techniques into the task.
When to use
- Proactively, at the start of any non-trivial implementation, before writing code.
- Choosing between libraries, algorithms, or architectural patterns.
- The user asks for "the best way", "SOTA", "how the top projects do this".
When not to: trivial edits, pure refactors with no new approach, or when the user has already specified the exact library/approach to use.
How to run
Dependencies are installed by install.py (just httpx). Run the bundled script
by its full path — it works from any working directory:
python "$HOME/.claude/skills/sota-scout/scripts/scout.py" \
--query "<concise description of what we're building>" \
--lang <github-language-if-relevant> \
--top 8 \
--json repos.json(scripts/scout.py is relative to this SKILL.md; use this skill's own directory.)
- Derive
--queryyourself from the task — use concise, canonical keywords (2-4 terms), not a sentence. GitHub repo-search ANDs terms and does not stem, so prefer the common noun form ("fastapi rate limit", not "rate limiting for my FastAPI app"). If results look thin or off-target, re-run with a variant of the key term ("rate limiter" vs "throttle", "vector db" vs "embeddings store"). - Set
--langto the project's language when it matters. - A GitHub token (
GITHUB_AUTH_TOKENin the repo.env) raises limits and enables more signals. Without it the script still runs, with a smaller pool — that's fine. - Flags:
--top N,--min-stars N(auto-relaxes if thin),--deep(extra enrichers),--no-technique(skip README fetch),--json PATH,--out PATH.
What you get
- stdout /
--out: a markdown brief — ranked repos with score, tier, the "why" (top contributing signals), README outline, and README/architecture excerpts. --json: structured results (scores, per-signal breakdown, technique excerpts).
What to do with it (don't stop at the list)
- Read the brief. For the top 3-5 repos, write a short technique card each:
- What's novel here (the core idea/algorithm/architecture).
- How it differs from the obvious/default approach.
- How to adopt it in our task — concretely, and whether it's adoptable in isolation or implies a larger change.
- Synthesise a recommendation: which technique(s) to use here and why.
- Then implement, building on those references. Cite the repos you drew from.
Treat the score as a way to order the shortlist, not a verdict — see
references/scoring.md for what each signal means and which are proxies.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

