Before anyone builds a new skill, plugin, or script, check whether one already exists. This skill fetches the current state of both tooling repos live — no stale catalog.
Run the following gh api commands in parallel to discover everything available.
Skills (embarkvetlabs/claude-configs):
gh api "repos/embarkvetlabs/claude-configs/contents/skills" --jq '.[].name'
For each skill name, read its SKILL.md frontmatter (name + description):
gh api "repos/embarkvetlabs/claude-configs/contents/skills/<name>/SKILL.md" --jq '.content' | base64 -d
Plugins (embarkvetlabs/embarkvet-claude-plugins):
gh api "repos/embarkvetlabs/embarkvet-claude-plugins/contents/plugins" --jq '[.[] | select(.name | startswith("_") | not) | .name]'
Skip any directory whose name starts with _ (e.g. _template). For each real plugin, read its README.md if present:
gh api "repos/embarkvetlabs/embarkvet-claude-plugins/contents/plugins/<name>/README.md" --jq '.content' | base64 -d
For skills inside a plugin (skip gracefully if skills/ subdirectory doesn’t exist — a 404 means this plugin has no skills):
gh api "repos/embarkvetlabs/embarkvet-claude-plugins/contents/plugins/<name>/skills" --jq '.[].name'
gh api "repos/embarkvetlabs/embarkvet-claude-plugins/contents/plugins/<name>/skills/<skill-name>/SKILL.md" --jq '.content' | base64 -d
For agents inside a plugin (skip gracefully if agents/ subdirectory doesn’t exist):
gh api "repos/embarkvetlabs/embarkvet-claude-plugins/contents/plugins/<name>/agents" --jq '.[].name'
gh api "repos/embarkvetlabs/embarkvet-claude-plugins/contents/plugins/<name>/agents/<agent-name>.md" --jq '.content' | base64 -d
Hooks (embarkvetlabs/claude-configs):
gh api "repos/embarkvetlabs/claude-configs/contents/hooks" --jq '[.[] | select(.name | endswith(".sh")) | .name]'
For each hook, read its header comment (first 20 lines is enough):
gh api "repos/embarkvetlabs/claude-configs/contents/hooks/<name>" --jq '.content' | base64 -d | head -20
Batch these calls aggressively — fetch all skill SKILL.md files in parallel, all plugin READMEs in parallel, etc. Do not fetch sequentially.
The user’s query is in the skill arguments. Scan everything fetched for tools whose description, trigger keywords, or stated purpose overlaps with the ask.
Cast wide — an 80% match is worth surfacing. Rank by fit: exact trigger-keyword match > description overlap > general category match.
For each match, output:
**<Tool name>** (`/<skill-name>` for skills · `<plugin>:<agent>` for agents)
Source: <repo> · <path>
<One sentence on why it fits this specific query.>
<Any caveats — things the tool does NOT cover, or extra steps needed.>
If multiple tools match, order best-to-worst and note if they should be combined (e.g. /plan-task then /build).
End with: “No need to build anything new — use the above.”
Say clearly that no existing tool covers this. Name the closest thing and why it falls short. Then stop — do not start building.