spec-initlisted
Install: claude install-skill ainova-systems/intelligence-dev-packs
# Initialize the Spec-Driven Docs Substrate
Stand up the repository's shared knowledge base - the layer both humans and agents read - so the rest of the spec pack (`spec-create`, `spec-execute`) has somewhere to write. Safe and additive: nothing is deleted, drafts are inferred-until-confirmed, the owner approves before anything is authoritative.
## Resolve the target structure (learn, then profile, then default) - never impose
1. **Learn from the project first.** If the repo already has a docs structure (`docs/`, `Documentation/`, a docs rule, sibling feature docs), **adopt it** - search first, read two sibling docs, mimic don't invent. A mature existing structure wins over the default; do not restructure what already works.
2. **Profile** pins paths when ambiguous: `specs_dir`, `features_dir`, `rules_dir`, `decisions_dir`, `spec_grouping`.
3. **Greenfield default** = the ai-first-docs tree (see `references/docs-tree.md`): `docs/` with `model.md`, `glossary.md`, `rules/R-NNN-*.md`, `features/<feature>.md`, `specs/NNN-<feature>/{NNN-requirements,NNN-plan}.md`, `decisions/` (date-named `yyMMdd-*.md` by default), `architecture/`, `guides/`, `_inbox/`.
## The layered model (what this skill owns)
Ordered by how strongly each constrains an agent: an agent follows an executable check more reliably than prose.
- **Layer A - executable constraints** (lint/boundary/contract-test configs, `tests/`): the strongest. **Out of scope here** - set up separately.
- **Layer B - machine-re