title: Find the skill files description: Built-in skills live under skills/ in the repository — personal, team-shared, or upstream, and, for jurisdiction- or practice-area-specific work, which upstream. url: https://legalquants.github.io/lq-ai/skills/where-skills-live/ checked-against: 671eab0d (2026-10-25) status: draft kind: explanation sources: docs/skills/where-skills-live.md, docs/adr/0012-db-backed-user-skills.md, docs/adr/0024-jurisdiction-and-practice-area-expansion.md, skills/CONTRIBUTING.md, api/app/api/user_skills.py, README.md Built-in skills are stored under `skills/` in the repository. Each has a directory containing `SKILL.md` and, where needed, a `reference/` folder or a `test-plan.md`. ## Read before using Look through the instructions and supporting files before you rely on a skill. Reference content is included in the AI request, so it can affect the answer and what information reaches a provider. ## Record the version you used Keep the skill revision with your test results. An author name or version label describes the file; it does not show that someone independently reviewed it. ## Personal skills A skill saved for your own use is a database record, not a change to the repository. Your version can take the place of a built-in with the same name in your own chats; other users keep their own version or the built-in. This is a useful starting point for adapting a task without publishing it. ### Details Save it through the Skill Creator or the skill wizard inside the running application; it lands as a row in the `user_skills` database table, scoped to you ([ADR 0012](https://github.com/LegalQuants/lq-ai/blob/671eab0d/docs/adr/0012-db-backed-user-skills.md)). It **shadows** a built-in of the same name in your own chats only — if you fork the shipped NDA Review skill and tweak its severity calibration, your version renders for you; everyone else, including the skill's original author, keeps seeing the built-in. Nothing here touches this repository or requires a PR. ## Team skills The API also supports team-owned skills, with team-admin checks for creating and changing them. These remain database records and belong in the database backup. The older design record deferred this feature, but the reviewed code includes it; do not rely on that old "deferred" label. ### Details The same wizard supports a team-scoped skill, visible to everyone on that team rather than to you alone — a team-admin action, gated in [`api/app/api/user_skills.py`](https://github.com/LegalQuants/lq-ai/blob/671eab0d/api/app/api/user_skills.py). [ADR 0012](https://github.com/LegalQuants/lq-ai/blob/671eab0d/docs/adr/0012-db-backed-user-skills.md) deferred team scope to a follow-on task (D8.1); the shipped code implements it — a `teams` table ([`api/alembic/versions/0014_create_teams.py`](https://github.com/LegalQuants/lq-ai/blob/671eab0d/api/alembic/versions/0014_create_teams.py)) and `owner_team_id` on `user_skills` — and no later ADR records the extension, so ADR 0012's "deferred" text is stale against the running code as of the checked commit. Still a database row, still no repository change — this is the right home for a skill your organization standardized on internally but has no reason to publish. ## Publishing a skill The curated built-ins live in this repository's `skills/` directory. The contribution policy routes community work-product skills to [`LegalQuants/lq-skills`](https://github.com/LegalQuants/lq-skills). Claim a task before starting and check that repository's contribution and license terms. Publishing a skill is separate from saving or sharing it inside your installation. For most skill contributions — a new jurisdiction's calibration for an existing review skill, or a new domain skill entirely — that's this route. Per [README.md §"Starter skills"](/lq-ai/operate/install-docker-compose/#starter-skills-ship-with-m1), this repository's `skills/` holds the fifteen first-party built-in skills — the ten M1 starter skills plus `case-law-research` and four table-mode/internal skills (`contract-snapshot`, `msa-snapshot`, `nda-snapshot`, `playbook-easy-extract`). The community repository is where jurisdiction- and practice-area-specific contributions accumulate, under review norms that mirror [`skills/CONTRIBUTING.md`](/lq-ai/skills/attestation/)'s bar (practicing-attorney attestation, peer review), but live in that repository, not this one. ### Details [ADR 0024](https://github.com/LegalQuants/lq-ai/blob/671eab0d/docs/adr/0024-jurisdiction-and-practice-area-expansion.md) settled the routing after several contributions stalled on exactly this question. An incoming jurisdiction or practice-area contribution is classified into one of four shapes: 1. **Skills (S2) go to `legalquants/lq-skills`.** Community work-product skills route to the community skills repo; this repository's `skills/` directory holds the curated first-party set. 2. **Authority sources (S1) go to this repository** as sources under ADR 0021, with the security review gateway-egress changes already carry. 3. **Corpora / statutory graphs (S3) go to org-level repos** per the PRD §9 DE-264 pattern; anything S3 that reaches operators is gated in this repository as an S1-class change. 4. **Practice areas excluded by PRD §1.6 (S4) require a mini-PRD and a committee-decided PRD amendment first**; the proposal's S1/S2/S3 parts route normally once the amendment lands. Litigation is chief among the practice areas PRD §1.6 excludes today. ADR 0024 also commits to a standing ledger at `docs/contribute/coverage-map.md`, checked before a contribution is routed and updated in the same sweep. **That file is not present in the repository as of the checked commit.** Until it exists, claim your contribution on a tracking issue first ([`skills/CONTRIBUTING.md`](/lq-ai/skills/attestation/)), same as any skill contribution, so two contributors don't duplicate the same jurisdiction or practice area. ## If the contribution adds more than instructions A new outside research source changes the application's network access and needs the security review gateway-egress changes already carry ([ADR 0021](https://github.com/LegalQuants/lq-ai/blob/671eab0d/docs/adr/0021-content-source-registry-and-free-source-expansion.md); CODEOWNERS routes `gateway/**` changes to security review), with the adversarial read in [docs/security/external-contribution-vetting.md](https://github.com/LegalQuants/lq-ai/blob/671eab0d/docs/security/external-contribution-vetting.md) for authors outside the known-contributor circle. A dataset may need a separate repository. A proposal outside the stated product scope needs a scope decision before implementation. ADR 0024 describes these routes (in detail above); ask a maintainer to classify a mixed proposal. ## Next - [What counts as a reviewed skill?](/lq-ai/skills/attestation/)