You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
skills: spoke files under context/, reference/ and references/ still carry a literal ${CLAUDE_PLUGIN_ROOT} that the Read tool never substitutes #5273
The Read tool does not substitute ${CLAUDE_PLUGIN_ROOT}. A spoke file (under a skill's context/, reference/ or references/) that tells the model to run ${CLAUDE_PLUGIN_ROOT}/scripts/x hands the model a literal, unresolved token. The path fails when the model runs it. #4613 reported this for claude-memory context/audit.md; it was fixed in claude-memory 0.13.5 (fix PR #5270 adds a gate). Other plugins were not swept. Related closed precedents fixed only their own plugins: #1691, #2237.
Count
As of origin/main aebb9bb1e, git grep -n CLAUDE_PLUGIN_ROOT -- 'plugins/*/skills/*/context/*' 'plugins/*/skills/*/reference/*' 'plugins/*/skills/*/references/*' finds 345 hits in 32 plugins. Highest: see the per-plugin list in the first comment. #4686's "recorded 18 other plugins" figure is stale. Some hits are prose or already have a fallback, so each needs triage (repo-hygiene was partly treated in #2237).
Pattern that works
claude-memory 0.13.5: SKILL.md renders <skill-dir> from ${CLAUDE_SKILL_DIR}, spokes cite <skill-dir>/scripts/..., with a dated verification record. The plugin-local gate plugins/claude-memory/scripts/spoke-script-paths.test.sh (PR #5270) is the template for a repo-wide check under scripts/.
Acceptance criteria
Every hit is triaged: fixed or justified (prose, has working fallback).
One check script under scripts/ gates the pattern repo-wide.
Activity
kyle-sexton commentedon Sep 29, 2026
Top counts at origin/main aebb9bb: source-control 71, discovery 36, knowledge 31, claude-ops 31, work-items 26, claude-config 21, repo-hygiene 18, disk-hygiene 15, review 12 (32 plugins, 345 hits total).
fix(claude-memory): repoint audit eval, correct 0.13.6/0.13.7 changel…
feat(scripts): gate skill spokes that name paths through the plugin-r…
fix(discovery): bump to 0.25.12 for spoke plugin-root fix
fix(discovery): resolve skill spoke paths without the plugin-root tok…
chore(knowledge): bump to 0.14.12 for the spoke path fix
fix(knowledge): write skill spoke paths as <skill-dir>, not the plugi…
fix(claude-ops): resolve skill spoke paths without the plugin-root token
fix(work-items): resolve skill spoke paths without the plugin-root token
fix(review): resolve skill spoke paths without the plugin-root token
fix(plugins): resolve skill spoke paths without the plugin-root token…
fix(plugins): keep three skills under the line cap and the pinned int…
docs(disk-hygiene): say which spokes define the engine directory in t…
docs(disk-hygiene): define the engine directory beside the catalog co…
fix(plugins): keep the literal token in prose that names it and basel…
fix(plugins): resolve skill spoke paths without the plugin-root token (…
fix(planning): hand-write the named-field reconfirm row so it round-t…
fix(discovery): record the explorer dispatch reproduction and add the…
fix(planning): resolve loop.md spoke paths without the plugin-root to…