fix(loop-cost): reject zero cadence intervals

[Iams4kura]

Summary

Reject zero-length cadence intervals so loop-cost fails clearly instead of emitting Infinity runs/day and null JSON estimates.

The parser previously accepted 0m as zero milliseconds, which made the runs-per-day calculation divide by zero.

Related issue

No existing issue or pull request matched the exact 0m symptom or the affected parseInterval path.

Changes

• New pattern or starter (followed templates/pattern-template.md + updated registry.yaml)

• Doc / example / story improvement

• Tool / CLI change under tools/

• Test only

• Other

• Validate that a parsed cadence is greater than zero before converting it to milliseconds.

• Add a regression test that rejects 0m.

• Keep the repository-tracked dist runtime synchronized with the TypeScript source.

Checklist (from CONTRIBUTING)

• Links work from README or the relevant index (no links changed)
• No secrets, tokens, or internal company URLs
• STATE.md* examples use .example suffix (no state examples changed)
• Safety-related content references docs/safety.md (no safety content changed)
• If you touched a package under tools/<pkg>: cd tools/<pkg> && npm ci && npm test
• If you touched the registry: root npm ci && npm run validate:registry (registry not changed; the repository validation gate passed)

Testing / Dogfood

• cd tools/loop-cost && npm test — 13 tests passed
• bash scripts/ci-validate-gates.sh
• bash scripts/ci-audit-gates.sh
• Manual review of generated state / skill output if scaffolding changed (no scaffolding changed)

Regression evidence

• Before: cd tools/loop-cost && npm test exited 1 because the new test did not observe an exception.
• After: cd tools/loop-cost && npm test exited 0.
• CLI behavior after the fix: --cadence 0m exits 1 with The interval value must be greater than zero.

Scope

• 3 files changed, +14 / -2 lines

---

Docs, stories, adopters, and small tests: maintainers aim to review within 48 hours (same-day when possible).

[cobusgreyling]

Zero cadence (0m) should be rejected. Check is cheap and the existing regex already only allows digits+unit; the extra value<=0 guard plus test is correct.