[OPS] Master-repo skill menu — list every skill synced from the .github father repo into this repo (grouped by category) and run the one Elie picks manually. Use when the user types /hq-skills or says "show me all skills", "list the skill menu", "what skills are available here", "pick a skill", or "apply a skill manually". Auto-apply by context stays the default; this menu is the manual override.
Installs into .claude/skills of the current project.
Are you the author of Hq Skills?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/edagher92-coder-hq-skills)
---
name: hq-skills
description: >
[OPS] Master-repo skill menu — list every skill synced from the .github
father repo into this repo (grouped by category) and run the one Elie
picks manually. Use when the user types /hq-skills or says "show me all
skills", "list the skill menu", "what skills are available here",
"pick a skill", or "apply a skill manually". Auto-apply by context stays
the default; this menu is the manual override.
---
# HQ skill menu
Present the full synced skill registry for THIS repo and run whichever
skill Elie selects. Never invent skills; the menu is exactly what exists
on disk.
## Steps
1. **Build the registry from disk.** Glob `*/SKILL.md` under the repo's
skills directory (`.claude/skills/` preferred; top-level `skills/` as
fallback). For each, read the frontmatter `name` + `description`, and
the `[CATEGORY]` tag if the description carries one. If the vendored
`skill_router` package is present, prefer
`python -m skill_router --registry` — same data, already parsed.
2. **Show the menu.** One numbered table grouped by category tag
(`[CONTENT]`, `[FINANCE]`, `[OPS]`, `[STRATEGY]`, `[REVIEW]`,
`[DESIGN]`, `[ADS]`, `[EFFICIENCY]`, untagged last), each row:
`N · /skill-name — one-line summary`. Keep summaries to one line; no
prose between rows.
3. **Run the pick.** When Elie answers with a number or name, invoke that
skill via the Skill tool with any arguments he gave. If the choice is
ambiguous, show the 2-3 closest rows and ask which.
4. **Delegation note.** If the chosen skill's work is then delegated to a
subagent or worker model, attach that SKILL.md's content to the
delegate's prompt — workers have no native discovery (see the
account rule "Skills pass down the model ladder").
## Rules
- Read-only over the registry: never edit, create, or delete skills here —
skill authoring belongs in the master repo (`edagher92-coder/.github`
→ `claude-defaults/skills/`), which syncs everywhere.
- If a listed skill needs an MCP/connector this repo doesn't have (Xero,
Canva, Meta token), still list it but mark it `(needs <connector>)`.
- No pagination games: if the registry is long, show it all — one screen
of rows beats a back-and-forth.