Do GSD subagents follow "read from absolute path" instructions in bridge SKILL.md files?

[anas-yousef]

Do GSD subagents follow "read from absolute path" instructions in bridge SKILL.md files?

I have globally-installed skills under ~/.agents/skills/ (installed via npx skills add on macOS), and project-specific skills under .claude/skills/ in my repo.

To avoid duplicating skill content, I created "bridge" SKILL.md files in my project that reference the global skills by absolute path. For example, .claude/skills/code-reviewer/SKILL.md contains:

# Code Reviewer

> Bridge to globally-installed skill.

## Load Full Skill

Read the complete skill and rules from:
- `~/.agents/skills/code-reviewer/SKILL.md`
- `~/.agents/skills/code-reviewer/rules/*.md` (if directory exists)

## Project-Specific Additions

- V2 naming compliance check
- i18n compliance (3 language files)
- RLS on new tables

I can confirm from the executor logs that GSD subagents do discover .claude/skills/*/SKILL.md files — I see them ls-ing the directory and reading into the skill folders.

My question: Is this bridge pattern sufficient? Will a gsd-executor or gsd-planner agent actually follow the instruction to Read the full skill from the absolute path (~/.agents/skills/...), or does it just read the local SKILL.md and stop there?

If this doesn't work reliably, should I instead:

1. Copy the full skill content into each project SKILL.md?
2. Use symlinks?
3. Some other approach?

Any guidance on the recommended way to make globally-installed skills available to GSD subagents would be appreciated. Thanks!