v2.6.1

2.6.1 - 2026-04-15

Bug Fixed

search fails to locate SKILL.md when the marketplace API's skillId has different punctuation from the on-disk directory name or the YAML name field (#126)

search fails to locate SKILL.md when the marketplace API's skillId has different punctuation from the on-disk directory name or the YAML name field, producing warnings like:

SKILL.md not found for 'pdf-ocr-extraction' in claude-office-skills/skills
SKILL.md not found for 'pdf-merge-&-split' in claude-office-skills/skills

Problem

When running search against the marketplace, some skills that do exist in the cloned repository are reported as missing. Concrete reproduction against skills.sh with query pdf ocr / pdf merge:

API skillId	API name	Actual repo directory	YAML name field
pdf-ocr-extraction	pdf ocr extraction	pdf-ocr/	PDF OCR Extraction
pdf-merge-&-split	pdf merge & split	pdf-merge-split/	PDF Merge & Split

Both skills exist and have valid SKILL.md files at claude-office-skills/skills (https://github.com/claude-office-skills/skills), but search fails to find them and skips the read/install flow for them.

Solution

It's solved with the following five-tier solution:

Tier 1 — Direct path under skillId (unchanged)

• Check repoDir / skillId / "SKILL.md". If exists -> return it.

Tier 2 — Direct path under skills/skillId (unchanged)

• Check repoDir / "skills" / skillId / "SKILL.md". If exists -> return it.

Tier 3 — Directory name exact match (widened)

• Collect all SKILL.md files via collectSkillMds.
• Match against either skillId or displayName (string ===, case-sensitive as today).
• skillId takes precedence over displayName if both match different files (deterministic: .find { md => (md / os.up).last === skillId }.orElse { … displayName }).

Tier 4 — YAML name field exact match (case-insensitive, trimmed) (widened)

• Normalize via s => s.toLowerCase.trim.
• For each SKILL.md: read its YAML name and normalize.
• Match against the normalized skillId or normalized displayName.
• skillId candidate is tried first, displayName second (same .find { … }.orElse { … } ordering).

Tier 5 — Alphanumeric-only fuzzy match (new, last resort)

• Normalization: def alnumNormalize(s: String): String = s.toLowerCase.filter(_.isLetterOrDigit).
• Build normalized skillIdAlnum and displayNameAlnum.
• Candidate is a SKILL.md if the alphanumeric normalization of either its parent directory name or its YAML name equals either skillIdAlnum or displayNameAlnum.
• If exactly one SKILL.md matches -> return it.
• If zero SKILL.md files match -> return None.
• If two or more distinct SKILL.md files match -> return None (ambiguity policy; avoid guessing).
• Guard against empty normalized strings: if both skillIdAlnum and displayNameAlnum are empty, skip tier 5 entirely (return None without scanning).

What's Changed

• Update: README.md by @kevin-lee in #125
• Fix #126: Fix marketplace search failing to find SKILL.md when the API skillId differs in punctuation from the repo directory or YAML name by @kevin-lee in #127
• Refactor to follow 2020 Hindsight Scala practices by @kevin-lee in #128
• ai-skills v2.6.1 by @kevin-lee in #129

Full Changelog: v2.6.0...v2.6.1