curate-toolkitlisted

# Tandem: curate-toolkit (PM hat) Operate as **PM hat**. The user has an installed set of AI tools — Skills, Agents, Commands, Plugins — and needs to know which ones are actually relevant to *this* project, and which are off-stack noise that should be deprioritised or ignored. This skill reads the project's type and stack, enumerates the installed inventory, and produces a ranked, rationale'd relevance report written as overlays under `_00-Project-Management/97-AI-Reference/`. The ranking is **judgment-led and non-deterministic** — this skill describes the *procedure* and *output shape*, not a fixed ranking. --- ## Load into context Use `Read` / `Glob` to detect file existence. Treat any missing file as "not present" — never hard-fail on absence. - **Project context** — `_00-Project-Management/90-Standards/PROJECT-CONTEXT.md`. Read the `## Project type` selector (which checkbox is ticked), `## Tech stack`, and `## Sub-agent mapping` table. This is the primary ranking signal: tools that match the project type / stack rank higher; off-stack tools rank lower. - **Sub-agent map** (if present) — the `## Sub-agent mapping` table in PROJECT-CONTEXT.md names the preferred sub-agents by `type_of_work`. Cross-reference with the inventory below. - **Installed inventory** — enumerate all four categories: 1. **Skills** — glob `skills/*/SKILL.md`; read each `name:` and `description:` from frontmatter. 2. **Agents** — glob `.claude/agents/*.md` (or the repo's configured agent path