feat(runneth-apps): seed customer-owned setup templates

[ioana (ioanahayman)]

Summary

• Bumps runneth-default to 0.4.0 with customer-owned setup templates for integrations and glossary definitions.
• Seeds /agent/brain/integrations/ with a folder explainer, reusable integration templates, and protocols for integration type, quirks, and usage-feedback capture.
• Keeps Slack as the only app-specific default seed because Slack is both a connected integration and a delivery surface.
• Seeds /agent/brain/glossary/ with templates for primary KPI, attribution, winner definition, ad states, naming conventions, funnel stages, voice rules, and channel roles.
• Adds package instruction guidance for checking the integration catalogue, saving durable org-specific integration preferences, asking one lightweight setup question when it would improve future work, and saving durable customer definitions into glossary files.
• Extends the package index validator to accept the new seed_file resource type from the sister Agent Builder PR.

Context

This is the Runneth Apps side of Phase 1.5 and is stacked on the Phase One default package PR: #141

Phase 1.5 installs the smallest useful customer-owned setup layer for the analyst use case. It seeds the place and shape for org-specific integration and glossary knowledge without pre-filling every customer brain with broad app-specific docs.

The split is intentional:

• Track A platform knowledge stays in protected Runneth/system guidance and registered app skills.
• Track B org-specific knowledge lives in customer-owned brain files created from the seeded templates when there is connection, usage, or customer-instruction signal.

Research Grounding

The transcript, Eric integration knowledge-layer plan, Reza knowledge-layer framing, Faye-style setup references, prod conversations, and Motion's Runneth VM brain all point to the same package shape: seed editable setup structure, then create app-specific or customer-specific maps from real signal. Customers commonly customize KPI definitions, attribution windows, source-of-truth preferences, naming conventions, reporting formats, and integration usage rules, so broad pre-seeded per-app placeholders would add clutter before the agent knows what matters for that org.

The Motion VM brain audit grounded the reusable pieces in existing setup files: Faye-style integration protocols for capabilities, activation, practical guides, quirks, and usage feedback; glossary templates for KPI, attribution, winner definitions, naming conventions, funnel stages, channel roles, and voice rules; and default user-brain routing that treats these definitions as customer-owned context. Motion-specific facts, org credentials, account playbooks, and heavy onboarding/sync skills are intentionally not copied into this default package.

The seeded files are not generic API documentation. They define stable customer-owned homes for source-of-truth rules, KPI definitions, attribution windows, naming conventions, important fields/folders/channels/reports, known gotchas, and reusable glossary definitions.

Sister PR

Requires the Agent Builder package manager and prompt support for brain-only seed_file resources: https://github.com/Motion-Creative/agent-builder/pull/2320

Validation

• GITHUB_EVENT_PATH=/tmp/runneth-fleet-approved-event.json node --test scripts/validate-runneth-package-index.mjs
• git diff --check

Notes

• This PR is stacked on feat(runneth-apps): add Runneth default package #141 and should be reviewed/merged after the Phase One package addition.
• Package index validation still requires the runneth-fleet-change-approved label because runneth-default is auto-installable.