backlog-researchlisted

# Backlog Research Once an idea has been triaged as NOVEL, the next question is: "is this actually buildable, and what would it take?" Doing that by hand means flipping between vault docs, the user's `~/.claude/` configuration, the relevant skill specs, and external prior art. This skill does the full pass in one session: it loads vault context, walks infrastructure dependencies, runs targeted external research, assesses feasibility on four dimensions, and writes a structured ideation brief at the canonical location. If the recommendation is PROCEED, it also generates a draft plan directory (`spec.md`, `tasks.md`, `manifest.json`) so a build session can start immediately. A per-session budget cap keeps research from spiralling. This skill operates on the **manifest-derived backlog** (Plan 81 SP13/SP14/SP15): pre-plan ideas live as `{paths.plans_root}/_inbox/<slug>.md` notes (funnel status `new / triaged / briefed`); in-flight plans are `{paths.plans_root}/<NN>-<slug>/manifest.json` (`lifecycle.status`). The unified pickup view `{paths.plans_root}/_backlog.md` is generated by `librarian:backlog-index` and is **librarian-owned (`writers_allowed: ["librarian"]`) — this skill never writes it**. Research mutates the inbox note's funnel status / the plan manifest's `status` + `disposition`; the next `librarian:backlog-index` regen re-renders the row. Curly-brace tokens (`{vault.root}`, `{paths.plans_root}`, `{paths.hooks_state}`, `{foundation_repo}`, etc.) resolve at runtime fro