Questions reviewers actually asked.

spec-kit is a methodology — a process for writing specs.SpecScore is a storage format for the specs you write, plus a CLI for querying and editing them. The two compose: spec-kit can produce SpecScore files, and a spec-kit extension that writes the SpecScore format is in flight.

Pick spec-kit when you need a process. Pick SpecScore when you need specs that AI agents can lint, query, and mutate without reading every file in the repo. Run both when you want both.

Similar in spirit. All three are skill libraries that guide AI agents through ideate → specify → plan → implement. The difference is mechanical, not philosophical.

SpecStudio Skills writes its artifacts in the SpecScore format (a defined Markdown + YAML structure) and dispatches the specscore CLI for queries, linting, and status changes. Superpowers and agent-skills mostly read and write files directly via skill prose.

Pick SpecStudio Skills when token consumption matters — the CLI returns a structured answer to "all open Features with unresolved questions" in one call, where a file-based skill reads N documents to answer the same thing.

Skills can read and write files. They cannot reliably:

• enforce referential integrity across the spec tree (links resolve, IDs unique, indexes stay in sync);
• return a filtered list of "open Features with unanswered questions" without reading every file;
• mutate the right status field at the right time on every artifact.

The CLI gives agents one clean call per task — the same wayjq gives you one clean call to query JSON. Skills alone cost more tokens, drift more often, and forget to update the index.

"Score" reads two ways, and both are intended.

Partitura. A structured score of requirements — the way a musical score structures parts, voices, and timing. Open a SpecScore tree and you should be able to read it the way a conductor reads a score: hierarchy, parts, references, all visible at once.

A literal quality grade. The CLI linter is the deterministic score (does the spec structure hold up?). The/score skillis the AI-judged score — an A–F readiness grade per artifact, with per-axis feedback for the author. The "Score" half of the name becomes a feature the tool produces, not just an aspiration.

A measured side-by-side comparison is in flight and will publish before the Habr post — same task ("what should I work on next?"), run twice: once with a vanilla agent reading docs, once with the same agent calling the SpecScore CLI. Token counts and time-to- answer for both.

Until those numbers land, the claim is structural: an agent asking the CLI for open Features and unresolved questions gets a small structured response. The same agent without the CLI reads N documents — N often in the dozens. The CLI surface is built precisely to skip that read fan-out.