The five highlighted skills
Five of these earn their load-bearing spot.
Each one I reach for weekly. Below: the mental model, where you'd actually use it, and why it stays on this page.
A RyOS Skill · Knowledge transfer
/capsule — hand off the why
Compress an agent's working memory into a portable handoff that the next agent can actually pick up. Decisions, open threads, fixtures, voice — four sections, each load-bearing.
Use for
- Switching projects mid-week without losing the why
- Onboarding a new agent (or teammate) onto in-flight work
- Compressing a long session before context ages out
Why
Compression beats summarization. Decisions + open threads + fixtures travel cleaner than transcripts.
A RyOS Skill · Packaging
/projector — package once, deploy everywhere
Build a custom agent as a portable package — system prompt, knowledge files, persona schema. Then target five surfaces from the same source: Claude Project, ChatGPT Project, NotebookLM, Gemini Gem, Code subagent.
Use for
- Building one assistant once, deploying to all five surfaces
- Handing a packaged agent to a teammate or client
- Versioning agents like software — one source, many targets
Why
Custom GPTs and Projects look different but share 90% of their guts. Package it once, ship it five times.
A RyOS Skill · Knowledge mining
/mine — graveyards → capsules
Point it at any knowledge source — Apple Notes, repos, PDF folders, podcast transcripts, Slack exports — and it scans, scores, condenses, and anchors. Output is a capsule the rest of your agents can consume.
Use for
- Turning a folder of PDFs into an agent-readable capsule
- Mining years of personal notes for actually-useful patterns
- Pre-processing a podcast or research corpus before deeper synthesis
Why
Most "AI knowledge" tools index your noise. /mine looks for signal — and discards the rest.
A RyOS Skill · Autonomous research
/research — closed-loop, council-synthesized
Persona-driven web research that scores its own findings and iterates until ready, then hands the corpus to a four-lens council for synthesis. Output is a synthesis.md — convergent findings, open tensions, what to do next.
Use for
- Pre-product validation that goes deeper than a Perplexity query
- Investor or market scans where surface search has known holes
- Building a synthesis a council can argue with — not a link list
Why
One-prompt research is shallow. Closed-loop scoring + council synthesis catches what one-pass queries miss.
A RyOS Skill · Output medium
/ink — when markdown can't hold the idea
Beautiful HTML for the things markdown can't hold. Multi-pane comparisons, SVG diagrams, annotated diffs, side-by-side variants — composed with a curated design system, not raw tags.
Use for
- Council outputs, research syntheses, planning docs handed to humans
- PRD-style briefs with multi-pane comparisons or SVG diagrams
- Deliverables too rich for markdown but too small for a deck
Why
Markdown is for thinking. Ink is for handoff. The composition system means the HTML actually looks composed — not just rendered.