Repository navigation
Conversation
Both failures are caught by `skills-ref validate`.
1. template/SKILL.md declares `name: template-skill` while its parent
directory is `template`. The spec requires the name to match the
parent directory. README.md already calls it "the **template-skill**
in this repository", and nothing in the tree references the path
`template/`, so the directory is the side that is wrong — renamed to
`template-skill/`, which also keeps the README accurate.
2. skills/claude-api/SKILL.md has a 1068-character description; the spec
caps it at 1024. Removed 48 characters of redundancy rather than
rewriting any trigger:
- `langchain_openai|` — in a `grep -E` alternation the earlier
`openai` branch already matches every `langchain_openai` occurrence.
- `Claude, Anthropic, ` — the clause immediately before it already
says "the prompt names Claude/Anthropic".
- ` in any form` — the parenthetical that follows enumerates the
forms, which is what the phrase was pointing at.
No trigger token, model glob, or provider name was dropped. The
description is now 1020 characters.
Spec: https://agentskills.io/specification#name-field
|
Still The two fixes are independent and I'm happy to split them. If only one is wanted, say which and I'll drop the other.
That second one is a judgement call about someone else's carefully tuned trigger text, so if you'd rather spend those 44 characters elsewhere, tell me where and I'll move the cut. The only thing that matters is landing under the cap. Both are reproducible with |
|
Additional data point supporting fix #2 (the 1068-char Because the Agent Skills spec mandates Reproduced on GitHub Copilot CLI 1.0.79 with the current upstream $ copilot skill list So the impact is broader than the Claude Code truncation noted in #881: on Copilot CLI the |
2 participants
Two skills in this repository fail
skills-ref validateagainst the spec this repository is the reference implementation for.1.
template/SKILL.md— name does not match its directoryWhich side is wrong: the directory.
README.md:65already advertises it as "the template-skill in this repository", andgit grep "template/"returns nothing — no file references the path. So renaming the directory totemplate-skill/keeps the name the README promises and breaks no links.2.
skills/claude-api/SKILL.md— description is 1068 characters44 characters had to go. Rather than rewrite any trigger, I removed 48 characters of redundancy:
langchain_openai|grep -Ealternation, the earlieropenaibranch already matches everylangchain_openaioccurrence — the alternative can never fireClaude, Anthropic,in any formNo trigger token, model glob, or provider name was dropped.
Fable,Opus,Sonnet,Haiku,`anthropic`,`@anthropic-ai`,`claude-*`,`us.anthropic.*`,`[1m]`, and every entry in both provider lists are untouched.Result: 1020 characters.
If you would rather spend the 44 characters somewhere else in that description, say so and I will move the cut — the only thing that matters is landing under the cap.
Found with AgentCompass, an offline static analyzer for AI-agent repo readiness. Both findings were verified by hand against the spec before opening.