Spatial context awareness. The mind seeing itself.
The ASCII map above is real. It's a snapshot from the session that built this site.
Hypercontext is a knowledge skill for Claude Code. When invoked via /hypercontext,
the agent introspects on its own session — what it's working on, what files it's touched, what tools it's used,
how much context runway remains — and renders it as a spatial ASCII map.
The map above is a real hypercontext snapshot from the genesis session (2026-01-29) — the session that built this website and the skill itself. The ASCII map is the actual output format. The hero image above it is Gemini concept art imagining what hypercontext might look like as a visual instrument.
No fabrication. Every thread, file, tool count, and decision is real. The map is only useful if it's honest.
Eight sections, each earned by introspection.
How much of the context window is consumed. Gradient bar: ▓▓▓░░░.
Estimated, labeled as such. Tells you when you're running low.
Message density over time: ▁▂▃▅▆▇█. Are you accelerating or decelerating?
Sparkline tells the story.
Major work streams as connected boxes. Status indicators: ✅ ⏳ ❌ 💡.
Arrows show which threads feed into others.
What's recent, what's stale. Pure recency model: ████ = hot,
░░░░ = cold. No hallucinated "importance" scores.
What was modified ◆ vs. read-only ◇.
Evidence trail. What did the agent actually touch?
Tool invocation counts as inline bars. Reveals working style: Bash-heavy? Read-heavy? Edit-centric? The histogram tells.
External systems contacted: ✅ alive, ⏳ pending, ░ not contacted.
What's the agent connected to?
What's been committed ✓ and what's still open ?.
The cognitive state, not just the code state.
Built on cartography. ASCII as information architecture.
Same data. Multiple lossy views. Different models.
Hypercontext is one view of session state — the ASCII terminal view. But the same data can be rendered as a visual radar (Gemini-generated), a compact briefing (for context-limit recovery), or a dense JSON (for programmatic consumption).
Each view is lossy. Each reveals something the others miss. That's the pylon pattern: parallel views of related context, in multiple formats and levels of detail. Even different models processing the same signal — Claude for ASCII structure, Gemini for visual intuition — produce orthogonal insights.
"Beating against yourself" across models and formats is how you evolve a vocabulary.
Concept art by Gemini 3 Pro — imagining what hypercontext data might look like beyond ASCII.
Context as landscape. Hot threads are volcanic highlands, stale context is frozen tundra. Files are settlements — fortified cities (modified) vs. villages (read-only). Trade routes = tool usage.
Brass-bezeled radar. Center = NOW. Concentric rings = time. Thread blips at angular positions. Tool sectors in color. Context remaining on the outer ring. A scientific instrument for metacognition.
Brass CRT at 3am. Phosphor green thread diagrams, file trees with status markers, context usage bar, system heat gauge. The hacker's metacognition display.
Swiss International Style. 53% in huge numerals. Threads by weight (bold = hot, light = cold). Tool histograms as character bars. Bloomberg-terminal density meets Vignelli.
Round 2 concept art: cross-pollinated aesthetics. The visual vocabulary sharpens.
Refined. Labeled volcanoes (Website Build, Visual Generation, Skill Authoring). Frozen tundra for stale context. Fortified cities = modified files. Trade routes labeled with trip counts. The definitive cartographic view.
Brass astronomical orrery. Sun = current task. Orbiting planets = workstreams. The "Visual Concepts" planet has 4 moons — one per Gemini instance. Context usage arc. Heat gauge thermometer. Mechanical gearwork. A scientific instrument for metacognition.
The dual-panel synthesis. Left: thread orrery radar. Right: territory minimap. Bottom: context load bar. Concept art imagining the ASCII data rendered as a visual instrument panel.
The ASCII map's closest visual relative. Three-column grid: threads with heat bars, files with ◆/◇ markers, tools with count bars. Decisions and questions at bottom. Concept art — the same data the terminal skill outputs, imagined as graphic design.
Concept art: the floating island of cognitive territory. Orrery, metrics panel, remaining context ocean. Three visual vocabularies fused.
Reading IS installing.
Hypercontext is a knowledge skill for Claude Code. No scripts, no dependencies. Place the SKILL.md in your skill directory and invoke /hypercontext to see your session state rendered as an ASCII map.
The map is not the territory. But a good map lets you navigate territory you can't directly see. An AI's context window is invisible territory. Hypercontext makes it visible. — genesis session, 2026-01-29
Self-awareness as a tool. The agent that can see its own state can reason about its own limitations. The context bar isn't just a progress indicator — it's metacognition rendered as ASCII. — design rationale
The map of the map-making. 12 visual concepts, 4 rounds, 4 aesthetic vocabularies, 1 synthesis.
Round 1 — Explore: Four divergent aesthetics generated in parallel. Terminal phosphor, cartographic territory, cognitive radar, Swiss typographic. Round 2 — Refine: Cross-pollinated the strongest elements. Territory map got labeled volcanoes, orrery got real workstream planets, the hybrid merged radar + territory, typographic got real data. Round 3 — Synthesize: Cinematic poster, brand icon (three rings), social card. Round 4 — Meta: This making-of image. A hypercontext map of the hypercontext visual development process. Recursive.
All visuals generated by Gemini 3 Pro (nano banana). Two instances running in parallel per round — "beating against yourself" to evolve vocabulary. The pylon pattern: same data, multiple lossy views, different prompts.
