import-existing-projectlisted

# /import-existing-project — bridge, don't move User ruling C-0068(a) (verbatim, the origin of this skill): *"there might be other people who developed their research projects in a different way and now want to switch to TrialError format... it can't be done programmatically for all of the possible projects, so we need a skill for claude code with an explicit instruction to use it to symlink files from non-TrialError research project to TrialError structured project."* The governing principle is **bridge, don't move**: the foreign project's own directory layout, git history, and (often huge) tool/model caches never move on disk. TrialError's `trialerror.toml` `[paths]` knobs and `[paths].ingest_roots` point INTO the existing tree instead, and only TrialError's own small derived output (chunks, embeddings, rendered views) lands under the new program root. Nothing about this procedure is mechanical end-to-end — every numbered step below is a judgment call you make WITH the user, not a script you run unattended. It genuinely cannot be done programmatically for an arbitrary foreign project (C-0068(a)'s own framing) — that is exactly why this is a skill (guided judgment) and not a CLI command (a fixed transform). This skill was distilled from a real import (a 40GB, multi-year, pre-TrialError research program with its own idiosyncratic layout) done this same way: fresh scaffold outside the foreign repo, then config-bridged rather than copied. The worked design doc for that speci