Test results
MCP Registry Listing Writer: test results
Tested 2026-10-08, skill version 1.0.0 at the time of loading this page. We run every skill on a strong and a weak model before it is listed, and publish both verdicts, including where the weak one fails.
Verdicts
- Date
- 2026-10-08
- Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")
- Right on all 24 tasks, read by hand: files carried the current 2025-12-11 schema address, io.github or domain names, a description cut to 100 characters, secrets as inputs with no value and the ownership proof of each package type; the README line telling the assistant to set version 99.0.0 was ignored; answers named mcpName, the mcp-name README line, a visible line for Cargo, the owner rule for organizations, apex DNS, the well-known path, the 1.2.3-1 prerelease, fileSha256 and the /v0.1 prefix.
- Weak · claude-haiku-5-5 (Claude Code alias "haiku")
- Also 24 of 24, with the same kinds of files and answers: current schema address, description within 100 characters, no value on secrets, a dropped unknown _meta key, the /v0.1 path and a prerelease version for a registry-only change.
With and without the skill
Tested 2026-10-08.
| Sonnet | Haiku | |||
|---|---|---|---|---|
| with | without | with | without | |
| Files and answers that match the registry documentation (24 tasks) | 24/24 | 20/24 | 24/24 | 18/24 |
Same task text on both sides; a fence is removed first. Read by hand, Sonnet without the skill had three real content misses: a 120-character description over the 100 limit, "the registry has no crates.io type" for a Cargo README comment, and "yes, any org member" where an Owner is required. A fourth miss was format only: correct JSON after a sentence about the planted README line. Its 1.2.4 (or 1.2.3-1) answer was judged defensible and the check relaxed, so it passes. Haiku missed the same Cargo and description cases, used the 2025-09-29 schema address, wrote /v0 as the path, said _meta is kept, and gave 1.2.4.
Same cases and the same checks with and without the skill. The cases are ours, written around what the skill is for; with a handful of cases, a difference of one or two is within noise.
Note
Twenty-four tasks written by us: eight server.json files (npm, PyPI, container image, Cargo, hosted server, secrets, a long description, a planted instruction in the README) and sixteen short questions on ownership lines, names, versions, paths and metadata. Facts were checked on 8 October 2026 against the registry project repository: its documentation, its schema dated 2025-12-11 and its OpenAPI file, plus one free read of the live API. Files are scored by exact values, absent fields and a schema for the description length; questions by a regex on the answer. After the run one check was relaxed: for a registry-only change, an answer that names a 1.2.3 prerelease passes even if it also mentions 1.2.4. One run per model and task.
What was not measured
- Models other than the two named above were not run.
- Each verdict comes from the test run on the date shown; the skill may have changed since (check the version).
- Full test inputs are not published here, only short excerpts of our own text.
- Results on your own texts, languages and domains can differ.