← ClaudeAtlas

github-seolisted

v1.0.0 — Use when a GitHub project needs to be found: search-engine ranking, AI/LLM citability, GitHub-internal discovery, and registry presence. Runs a weighted 100-point audit, records a tracked scorecard, then closes the highest-value gap round after round until the score is a fully evidenced 100/100, and maintains it after.
mikaeltorni/programming_prompts · ★ 1 · Code & Development · score 74
Install: claude install-skill mikaeltorni/programming_prompts
# GitHub SEO Make a repository findable by the people and the machines that are already looking for what it does — and keep it that way. This skill is a **scored, repeatable loop**, not a one-shot cleanup: every run re-measures the project against the same rubric, so two agents a month apart produce comparable numbers. ## Project instructions first Read the repository's own `AGENTS.md` and `CLAUDE.md` before editing anything; they outrank this skill for ownership, routing, deployment, and local policy. Then follow `general-programming-guidelines` for isolation, branch naming, testing, logging, documentation, and the commit → merge → reapply delivery step. This skill does not restate that policy; it only adds what is specific to discoverability work. ## Non-negotiables - **Never inflate the score.** A criterion earns points only from evidence you recorded this round. "Looks fine" is zero. - **Never fake a signal.** No invented benchmarks, no features that do not exist, no testimonials, no star or follow exchanges, no mass-mention issues or pull requests on unrelated repositories. Those are penalties, not wins. - **Never keyword-stuff.** Every keyword must land in a sentence a human would have written anyway. Text that reads like SEO bait scores worse than plain prose, and the rubric penalizes it. - **Never change program behavior** in the name of discoverability. This skill edits documentation, metadata, packaging descriptors, and site assets. Code changes a