release: v0.3.0 — verifiability and scoring standards

[nonlinearnishu]

Release v0.3.0. Version bump (package.json + VERSION) committed as 658d0aa; annotated tag v0.3.0 already points at that commit.

Merge with a merge commit (not squash) — the v0.3.0 tag references 658d0aa, so that SHA must remain in main's history or the tag dangles.

---

Outcome: v0.3.0 — Created first pass at verifiability and scoring standards. This creates a banded conformance score, a verification subagent, and a delta view. This is wired up during each PR through a GitHub Action.

Why: The kit could lint conventions but couldn't measure how well a project conformed, or tell a contributor which evals must run for a given change. This adds the four-tier verification spectrum, a severity-aware conformance score, and a PR-boundary gate.

26 files changed, +1099 / −47

feat:

• Adding verifiability and scoring standards to the kit
• add factory-verification subagent and conformance gate configuration options

ci:

• dogfood the conformance gate on factory-kit itself

Surface: Skills ~factory-ci.md ~factory-llm-workflows.md ~factory-pitfalls.md +factory-verification.md · Agents +verification-engineer.md · Bin ~factory-kit-check.js ~factory-kit.js

🤖 Generated with Claude Code

[github-actions]

🏭 Factory conformance — ✅ Pass

This PR: ✅ no new criticals

Severity	Repo	This PR
Critical	0	0
High	0	0
Medium	0	0
Low	0	0

Coverage: 6/48 pitfalls machine-checked (13%).

⚠️ 5 critical-class pitfalls are not machine-checked — a passing grade does not clear them: untested-rollback-shipped-to-prod, prod-destructive-write-no-layer-c-snapshot, llm-auto-runs-runbook-commands, phi-in-email-no-baa-check, ai-generated-code-no-review-queue.