SKILL.md YAML block scalar description not parsed correctly

[zlh124]

Description

When a SKILL.md frontmatter description field uses YAML block scalar syntax (> folded or | literal), the parser stores the indicator character (e.g. >) as the description value and ignores the indented content entirely.

For example:

description: >
  This is a multi-line
  skill description

The parsed description becomes ">" instead of "This is a multi-line skill description".

Steps to reproduce

1. Create a SKILL.md with a block-scalar description:

   ---
   name: test-skill
   description: >
     line one
     line two
   ---
   This is the skill body

2. Run /skills in DeepSeek TUI to list loaded skills, or load the skill via load_skill

3. Observe the description displayed for this skill — it shows > instead of the actual text

Alternatively, call SkillRegistry::parse_skill directly in a unit test to reproduce.

Expected behavior

• > (folded): content lines joined with spaces → "line one line two"
• | (literal): content lines joined with newlines → "line one\nline two"
• >- / |- (strip chomp): trailing blank lines removed
• >+ / |+ (keep chomp): trailing blank lines preserved
• Single-line descriptions (non-block-scalar): unchanged backward compat

Actual behavior

Block scalar continuation lines are completely ignored. The description field is set to the raw indicator string (e.g. ">").

Impact

Skills authored with multi-line descriptions (using YAML block scalars) show incorrect description text in /skills, load_skill tool output, and the system prompt's skills context block.

Fix

A fix has been committed to feat/skill-yaml-block-scalar branch (fork at zlh124/DeepSeek-TUI). It implements proper YAML block scalar parsing with indentation-based block collection, chomping (strip/clip/keep), folding (space-join) and literal (newline-join) modes, plus 13 unit tests covering all variants.

Environment

• DeepSeek CLI version: 0.8.40

[zlh124]

This is not only a display issue — SkillRegistry::parse_skill returns the incorrect description value in the Skill struct, so every consumer (the model prompt context, load_skill tool output, /skills command, and all skill listing commands) receives the wrong data. The fix is in the parser, not in any rendering layer.

[Hmbown]

Thanks @zlh124. This is a parser-level bug, not just display polish, because the wrong description reaches /skills, load_skill, and model prompt context.

PR #1908 is the candidate fix and is currently green across lint/tests. I am milestoning this to v0.8.44 for maintainer review rather than closing it before the PR lands.

[zlh124]

Thanks @zlh124. This is a parser-level bug, not just display polish, because the wrong description reaches /skills, load_skill, and model prompt context.

PR #1908 is the candidate fix and is currently green across lint/tests. I am milestoning this to v0.8.44 for maintainer review rather than closing it before the PR lands.

谢谢你，鲸鱼兄弟！