Portfolio: signalstudio.io · Contact: editorial.operating.system@gmail.com
I build AI infrastructure for editorial and marketing teams — document-grounded agent systems with modular skill routing, persistent learning loops, and structured eval frameworks. Three Claude plugins published in the Claude Code plugin ecosystem, plus The Gate, a production editorial quality-gate API.
Twenty years building design and brand systems at scale: Design Director at WIRED, Sr. Design Manager at Amazon Worldwide Communications, Associate Creative Director at the LA Times. At Amazon I built the brand system for an 800+ person global communications organization — brand guides, component libraries, and workflow infrastructure that ran without me in the room. That's still the problem. AI just changed the surface it's built on.
All four projects share one pattern:
Documents (Google Drive)
↓
client-context skill ← reads, structures, and caches brand context
↓
Specialist skill routing ← loads the right domain skill per command
↓
Output + eval layer ← scores output against a rubric before delivery
↓
Learning loop ← results logged back to Drive; each run reads prior learnings
Document grounding → specialist routing → eval layer → compounding learning loops. The Gate implements the eval layer as a standalone API callable by any pipeline.
Full diagram and design decisions: ARCHITECTURE.md
Editorial OS — Editorial Strategy System
Claude plugin MCP Google Drive Markdown skill architecture
Seven commands, five specialist skills, full editorial pipeline from audit to social repurposing. Point it at a Google Drive folder and your documents become the configuration — no setup forms, no brand questionnaires.
The Gate — Editorial Quality Gate API
TypeScript Express 5 PostgreSQL Drizzle ORM OpenAPI 3.1 Claude
A REST API that evaluates any draft against a fixed eight-criterion rubric and returns a machine-readable verdict: APPROVED, APPROVED_WITH_NOTES, REVISE, or BLOCKED — with 1–5 criterion scores, evidence quotes from the draft, a weighted confidence score, and specific revision instructions when action is required. SHA-256 hashed API keys, per-key rate limiting, PostgreSQL review history.
→ The eval layer, externalized. The rubric is fixed. The model cannot freestyle.
Email Marketing Manager — Newsletter Production System
Claude plugin MCP Google Drive Learning loops
Drafts newsletters from real documents, generates segment variants, predicts performance, and writes learnings back to Drive after every send. Institutional knowledge compounds in plain markdown any team member can read.
Pinterest Marketing Strategist — Pin Production System
Claude plugin MCP Google Drive Cloudinary
Composed pin batches from an approved image library and brand context. Tracks performance by image type × title pattern × board, and applies learnings to every subsequent batch.
Composer — Story Packaging System
Claude plugin Google Drive Rights metadata
Packages finished stories for publication: matched visual assets, usage rights read from embedded metadata, platform-ready crops, archival context. One brief in, one publishable package out.
### [Photo Editor](https://github.com/EditorialOS/photo-editor) — Post-Shoot Metadata Pipeline
`Claude plugin` `ExifTool` `XMP/IPTC` `Deterministic CLI`
Embeds rights, products, and shoot context into image files at ingest. Intelligence at intake, determinism at embed — the model never sits in the write path. Closes the loop with [Composer](https://github.com/EditorialOS/composer): rights written here are read at packaging.
### [Context-API](https://github.com/EditorialOS/Context-API) — Agent Memory Layer
`MCP server` `PostgreSQL` `OpenAPI` `Multi-tenant` `Proprietary`
The memory and brand-context layer the plugin suite's learning loops stand on. Multi-tenant MCP server with 39 tools auto-generated from a live OpenAPI spec — zero drift between API and tool surface — with per-workspace isolation and a scored eval harness (store → recall → isolation). Source-available, proprietary license.
The-Gate-VPS — Eval Engine, Deployed
MCP server Docker Multi-tenant Proprietary
The production deployment layer for The Gate: a multi-tenant MCP server exposing the eval engine as agent-callable tools. Ships a scored eval harness with pass-rate reporting and a working draft→eval→revise agent example.
## Install
/plugin marketplace add EditorialOS/editorial-os
/plugin install editorial-os@editorial-os
/plugin marketplace add EditorialOS/email-marketing-manager
/plugin install email-marketing-manager@email-marketing-manager
/plugin marketplace add EditorialOS/pinterest-marketing-strategist
/plugin install pinterest-marketing-strategist@pinterest-marketing-strategist
Runs in Claude Code. After adding the marketplace, install each plugin individually.
Portfolio: signalstudio.io · editorial.operating.system@gmail.com