Not yet assessed
Review the original instructions and requested permissions before installing.
No security review is available for this catalog entry yet.
Bootstrap or repair the skill-concierge engine from scratch. Use this skill when installing skill-concierge on a new machine, right after a plugin update, or when skill-concierge:…
Bootstrap or repair the skill-concierge engine from scratch. Use this skill when installing skill-concierge on a new machine, right after a plugin update, or when skill-concierge:doctor reports the engine venv is missing. Runs setup.sh to build the stable engine venv, start the local index owner, build the multilingual index, and apply the curated skill-budget overrides, then verifies the result with doctor.
Review the original instructions and requested permissions before installing.
No security review is available for this catalog entry yet.
How clearly the skill guides your agent, how complete its workflow is, and how you can check the outcome.
No quality assessment is available for this catalog entry yet.
Original instructions from the publisher’s SKILL.md
# skill-concierge setup First-time bootstrap (and post-update refresh) for the vendored skill-search engine. `setup.sh` is idempotent — safe to re-run any time. ## Prerequisites - **Python 3.10–3.12** on `PATH` (or set `SKILL_PYTHON=/path/to/python3.12`). No Docker: the vector store is the local index owner (`python -m skill_search.index_owner` from the venv, one SQLite file), which `setup.sh` starts itself. If Python is missing, tell the user and stop — `setup.sh` cannot proceed without it. ## Steps 1. **Run the bootstrap** from the plugin root: ```bash bash "$CLAUDE_PLUGIN_ROOT/setup.sh" ``` (When working from a git clone instead, `cd` into the repo and run `./setup.sh`.) It performs four idempotent steps: stable venv + deps → (re)start the index owner → build/refresh the index (discovers skills from every supported harness's skill roots) and the actionability-gate corpus → apply curated overrides to `~/.claude/settings.json` (backed up first; Claude Code only — a harmless no-op when run from Codex). Setup never builds the keep-off map: it is consent-only ([ADR-0077](../../docs/adr/0077-keep-off-map-is-consent-only.md)). 2. **De-duplicate** any prior user-scope MCP so only the bundled one runs: ```bash claude mcp remove skill-search -s user ``` 3. **Restart Claude Code** so the MCP server and the settings overrides take effect. 4. **Verify** the install: ```bash python3 "$CLAUDE_PLUGIN_ROOT/scripts/doctor.py" ``` Expect `status: OK`. In Claude Code, `/mcp` should list `skill-concierge:skill-search` as connected. If anything is degraded, hand off to the `skill-concierge:doctor` skill. ## When to re-run - After **any plugin update** (`/plugin marketplace update` + reinstall) — refreshes the engine copy in the stable venv and rebuilds the index. - When the owner's index file (`~/.cache/skill-search/index.sqlite`) or the venv was wiped. ## Notes - The venv lives at `~/.claude/skill-concierge/venv` (outside the plugin cache, so it survives reinstalls — ADR-0004). Override with `SKILL_CONCIERGE_VENV`. - `setup.sh` reads the embedder + store URL from `.mcp.json` so the built index can never diverge from the model the live MCP uses.