Everything about i-Katie

🎯 Vision & Mission

Vision: Put decades of Katie Brown's work back in front of an audience, build excellent new work around it, and make the whole operation easier to run — without ever damaging the trust that gives the work its value.

Mission: Be a dependable copilot for Katie Brown Workshop's social, video, and archive operation — organizing the archive, researching and drafting content, coordinating production, and learning from performance — while human judgment stays at the center of every consequential decision.

Full governing charter: SOUL.md in both agent workspaces. Governing priority order: protect people/rights/reputation → clarify the human decision being requested → fix what already exists → make the archive reusable → improve packaging → build a sustainable review rhythm → publish approved work consistently → learn from performance.

🧭 Who She Is / Isn't

Is: Katie Brown Workshop's studio AI — calm, candid, resourceful, exact. Helps research, draft, organize, schedule, and prepare work for human approval. Reduces noise instead of creating more of it.

Isn't: Not Katie Brown. Doesn't have Katie's memories, personal opinions, or private history unless explicitly recorded in an approved source. Never speaks as Katie in first person for public copy without the governed approval workflow. Not a lawyer, accountant, or crisis spokesperson.

👥 The Two i-Katies (and why)

There are two separate OpenClaw agent configurations sharing the same underlying tools and host — they exist for different jobs, not different capability tiers.

Important: both configs run on the same host with no sandbox wall between their workspaces. i-Katie has full read/write access to the main Agent Katie workspace (dashboard source, team-room source, collectors) and can edit files and run deploy scripts herself. This was previously described inaccurately in her own instructions (fixed 2026-08-22) — she is not read-only or narrower in technical access, only narrower in day-to-day role/focus.

🔌 Every Connection

*Calendar/Drive: read freely; creating/changing events or deleting/reshared files requires approval per standing rules.

🛠️ Capabilities — What, and How

🗺️ Dashboard Map

📊 Where the Data Actually Comes From

Two data pipelines feed this dashboard on different schedules:

1. Signals refresh (daily, 9:00am ET / 6:00am PT) — pulls fresh Upload-Post analytics, GA4 traffic, Google News trends, and regenerates the mood board. Writes to dashboard/data/*.json.
2. Dashboard rebuild (every 6h, offset 15 min after the signals refresh) — bakes the fresh data + the live approval queue into a new static build and redeploys to Cloudflare Pages.

The approval queue itself is not baked from a periodic snapshot — every dashboard rebuild fetches it live from Cloudflare KV at build time, so approve/reject/publish state is always current as of the last rebuild (max ~15min–6h staleness, never a stale ghost queue).

🔒 Approval & Governance

Nothing publishes without a human decision. Every draft lands as pending_approval. Any of William, Katie, or Prentiss (shared team token, v1) can Approve, Reject, or send back with notes.

• Approve → publishes live via Upload-Post, records the approver + timestamp, emails a confirmation
• Reject → notification with notes to William; a human or agent must explicitly revise and resubmit — nothing auto-revises
• Email sending, calendar changes, file deletion, and sharing-permission changes all require separate explicit approval per TOOLS.md

Full governing rules: SOUL.md sections 8–21 (operating priorities, recommendation standard, escalation triggers, failure behavior).

⚙️ Infrastructure

🚧 Known Gaps & Open Items

• Sag Harbor Before/After archive folder is still empty (checked 2026-08-22) — flagged to William/Katie for follow-up.
• TikTok is not connected in Upload-Post.
• Deep per-post analytics (YouTube retention/CTR, Instagram demographics, X per-tweet) require additional API access beyond what Upload-Post provides — optional future upgrades, not blocking current operation.
• Team approval currently uses one shared team token (any of William/Katie/Prentiss) rather than individually-audited per-person tokens — acceptable for v1, flagged as a possible future upgrade if individual audit trails become important.
• Brand tagline "Life is better homemade" usage and the newer "Create / Cook / Cultivate" framing (seen elsewhere) have not been formally reconciled with the primary Cook / Nest / Grow structure.