fix: top-level skill dirs so Claude discovers unprefixed names (#761)

* fix: top-level skill dirs so Claude discovers unprefixed names

Replace directory symlinks (gstack/qa → qa) with real directories
containing a SKILL.md symlink. Claude Code auto-prefixes skills nested
under a parent dir symlink, so /plan-ceo-review became "Unknown skill"
even with skill_prefix=false. Real dirs fix this.

Also syncs package.json version to match VERSION file and updates
test assertions to match the new mkdir + ln approach.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* docs: update symlink references to new top-level directory pattern

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* test: regression tests for top-level skill directory structure

Verifies the invariant that setup/relink creates real directories (not
symlinks) at the top level, with SKILL.md symlinks inside. This prevents
Claude Code from auto-prefixing skills with gstack- when using --no-prefix.

Tests added:
- unprefixed skills must be real dirs with SKILL.md symlinks
- prefixed skills must also be real dirs with SKILL.md symlinks
- old directory symlinks get upgraded to real directories
- cleanup functions handle both old symlinks and new dir pattern
- link function removes old directory symlinks before mkdir

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* test: namespace isolation tests for first install + mode switching

Verifies the core invariant: when you pick a prefix mode, ONLY that
mode's entries exist. Zero pollution from the other mode.

- first install --no-prefix: only flat names, zero gstack-* leaks
- first install --prefix: only gstack-* names, zero flat leaks
- non-TTY defaults to flat names
- switching prefix→no-prefix removes ALL gstack-* entries
- switching no-prefix→prefix removes ALL flat entries

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* feat: upgrade migration system — versioned fix scripts for broken state

Adds gstack-upgrade/migrations/ directory with version-keyed bash scripts
that run automatically during /gstack-upgrade (Step 4.75, after ./setup).
Each script is idempotent and handles state fixes that setup alone can't
cover. First migration: v0.15.2.0.sh runs gstack-relink to fix stale
directory symlinks from pre-v0.15.2.0 installs.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* test: migration script validation + v0.15.2.0 end-to-end fix test

Tests that migration scripts are executable, parse without syntax errors,
follow the v{VERSION}.sh naming convention, and that v0.15.2.0 actually
fixes stale directory symlinks by converting them to real directories.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* docs: upgrade migration guide in CONTRIBUTING.md + CLAUDE.md pointer

CONTRIBUTING.md: new "Upgrade migrations" section documenting when and
how to add migration scripts for broken on-disk state.

CLAUDE.md: added note under vendored symlink awareness pointing to
CONTRIBUTING.md's migration section when worried about broken installs.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

CHANGELOG.md

@@ -9,6 +9,10 @@ You can now run `/design-html` without having to run `/design-shotgun` first. Th
 - **`/design-html` works from any starting point.** Three routing modes: (A) approved mockup from /design-shotgun, (B) CEO plan and/or design variants without formal approval, (C) clean slate with just a description. Each mode asks the right questions and proceeds accordingly.
 - **AskUserQuestion for missing context.** Instead of blocking with "no approved design found," the skill now offers choices: run the planning skills first, provide a PNG, or just describe what you want and design live.
 
+### Fixed
+
+- **Skills now discovered as top-level names.** Setup creates real directories with SKILL.md symlinks inside instead of directory symlinks. This fixes Claude auto-prefixing skill names with `gstack-` when using `--no-prefix` mode. `/qa` is now just `/qa`, not `/gstack-qa`.
+
 ## [0.15.0.0] - 2026-04-01 — Session Intelligence
 
 Your AI sessions now remember what happened. Plans, reviews, checkpoints, and health scores survive context compaction and compound across sessions. Every skill writes a timeline event, and the preamble reads recent artifacts on startup so the agent knows where you left off.