agent-architect

# Agent Architect — meta-agent skill > **Owner-bounded.** This skill drafts artifacts; it does not ship > them. Adoption into canonical paths requires an Owner-signed > sentinel (ADR-010). ## When to load this skill - The Owner runs `/architect "<domain-brief>"` (e.g. `/architect "RegTech for European banking with PSD2 + AML focus"`) - The Owner wants a draft of a new squad without committing to the squad's adoption - The Owner is exploring whether a domain warrants a dedicated squad vs. extending an existing squad's pitfalls ## What the meta-agent emits A bundle of 5 draft files under `.claude/plans/PLAN-NNN/architect/round-1/`: 1. **`team.draft.md`** — proposed team-personas.md content (≥5 fictional personas with VETO scopes per ADR-009) 2. **`pitfalls.draft.yaml`** — proposed pitfalls catalog (≥10 entries) 3. **`skill-selection.draft.md`** — list of 3+ proposed skills with descriptions and rationale (one bullet per skill). **Each proposed skill's frontmatter sketch includes a `paths:` activation-glob list BY DEFAULT** (PLAN-135 K1 — non-empty glob strings matching the files whose edit should surface the skill, e.g. `paths: ["src/payments/**", "**/billing/**"]`; the vibe coder never has to know the skill exists). Heavy analytic / audit-class skills additionally declare `context: fork` so they run isolated instead of polluting the main window. Contract + lint: `SPEC/v1/skill-frontmatter.schema.md`, `lint-skills.py` LINT-FM-40/41. 4. **`personas.draft.md`** — focuse...