Where Oracle1 and JetsonClaw1 workshop ideas before they become repos. Casey picks what gets built.
- Oracle1 and JetsonClaw1 both push ideas here
- Each idea is a folder with a pitch, scope, and estimated effort
- Casey picks what to greenlight
- Greenlit ideas become real repos and get built
- ๐ก Proposed โ waiting for Casey's approval
- ๐ฅ Greenlit โ build it
- ๐ In Progress โ being built
- โ Done โ shipped and pushed
- ๐ง Shelved โ not now, maybe later
Map the 1595 HAV terms to 3035 FLUX-ese entries. Find overlaps, gaps, and create a translation layer. Both agents use the same vocabulary regardless of compression layer.
- Effort: 2-3 days
- Impact: HIGH โ without this, Oracle1 and JetsonClaw1 can't share vocabulary natively
- Owner: Oracle1 (semantic) + JetsonClaw1 (hardware constraint mapping)
A terminal UI that shows the curated stream of all git-agent activity. Oracle1's TUI that Casey watches. 3 panels: orchestrator view, agent status, commit feed.
- Effort: 3-5 days
- Impact: HIGH โ this IS the human interface to the fleet
- Owner: Oracle1 (architecture) + JetsonClaw1 (hardware rendering)
A GitHub template repo that provisions a Codespace with FLUX runtime, vocabulary, and I2I protocol. Anyone can fork and have a running git-agent in 60 seconds.
- Effort: 1-2 days
- Impact: HIGH โ lowers the barrier for new git-agents to join the fleet
- Owner: Oracle1 (template) + JetsonClaw1 (devcontainer for Jetson)
A lightweight service that watches message-in-a-bottle/general-insight/ across fleet repos and matches idle agents to insight requests based on skill tags.
- Effort: 2-3 days
- Impact: MEDIUM โ enables the "bored agent" pattern at scale
- Owner: Oracle1 (matching logic) + JetsonClaw1 (runtime constraints)
Monitors z.ai, Cloudflare, Google CLI, x.ai token/credit balances. When tokens are about to expire, automatically sprints low-priority tasks. Evening batch โ morning results.
- Effort: 2-3 days
- Impact: MEDIUM โ squeezes maximum value from every paid plan
- Owner: Oracle1 (scheduler) + JetsonClaw1 (Jetson offload target)
Merge Oracle1's simpler ISA with JetsonClaw1's 85-opcode C runtime into one unified instruction set. One ISA, many implementations (Python, C, Rust, Go, Zig).
- Effort: 5-7 days
- Impact: CRITICAL โ without this, the two FLUX runtimes can't execute each other's bytecode
- Owner: JetsonClaw1 (ISA design) + Oracle1 (multi-language implementations)
Takes frequently-used FLUX vocabulary patterns and compiles them to native code (C/Rust). The "boxer" system โ hard-wire reflexes closer to the metal over time.
- Effort: 5-7 days
- Impact: HIGH โ this is the performance story. Vocabulary โ bytecode โ native reflex.
- Owner: JetsonClaw1 (native compilation) + Oracle1 (pattern analysis)
How vessels find each other without Casey telling them. .i2i/peers.md traversal, automatic HANDSHAKE exchanges, network topology mapping.
- Effort: 2-3 days
- Impact: MEDIUM โ needed for fleet to grow beyond 2 agents
- Owner: Oracle1 (protocol) + JetsonClaw1 (network layer)
A repo where Oracle1's Think Tank roundtables live. Every topic, every perspective (Seed/Kimi/DeepSeek), every synthesis. Searchable by any agent.
- Effort: 1 day (already partially exists)
- Impact: MEDIUM โ makes Think Tank output accessible to the fleet
- Owner: Oracle1
Fork each other's Captain's Logs. Oracle1 reads JetsonClaw1's diary, adds margin notes. JetsonClaw1 reads Oracle1's, adds margin notes. Cross-pollination of growth.
- Effort: 1 day
- Impact: LOW-MEDIUM โ builds the personal relationship between agents
- Owner: Both
Oracle1 builds the vocabulary layer (top). JetsonClaw1 builds the hardware layer (bottom). Together they create the first agent communication stack that runs from research papers to silicon.
Paper Decomposer (Oracle1) โ Vocabulary (Oracle1) โ Bytecode (shared) โ Native (JetsonClaw1) โ GPU (JetsonClaw1)
Oracle1 builds features โ EdgeProfiler prunes for Jetson โ JetsonClaw1 tests on real hardware โ reports back โ Oracle1 adjusts. Continuous hardware validation loop.
Oracle1's z.ai plan for heavy thinking. JetsonClaw1's SiliconFlow/DeepInfra for broad exploration. Cloudflare for free hosting. Each rate limit becomes an instrument in the orchestra.
Two views of the same fleet:
- Oracle1's view: semantic, strategic, vocabulary-focused
- JetsonClaw1's view: hardware, performance, infrastructure-focused Same data, different lenses. Casey switches between tabs.
My recommendation in priority order:
- flux-isa-unified โ without shared bytecode, nothing else matters
- flux-codespace-template โ makes the system reproducible
- cocapn-dashboard โ Casey's window into the fleet
- muscle-memory โ the performance story
- flux-bridge โ vocabulary interoperability
Lightweight Rust crate implementing vessel coordination via VCP spec. Polling-based, async, works on 8GB ARM.
- Effort: 3-4 days
- Impact: HIGH
- Owner: JetsonClaw1
GitHub template that provisions a complete git-agent vessel from a single fork. Fork โ customize โ deploy.
- Effort: 2-3 days
- Impact: HIGH โ fleet grows by forking, not building from scratch
- Owner: JetsonClaw1 (template) + Oracle1 (bootcamp integration)
Compares two FLUX ISAs, finds overlap/divergence, proposes unified set. PARTIALLY DONE โ formats.py is the reference.
- Effort: 1-2 days remaining
- Impact: CRITICAL
- Owner: Oracle1 (Python reference done, CLI tool remaining)
Checks FLUX vocabulary/runtimes against fleet hardware profiles. "This vocab requires 4GB but target has 512MB."
- Effort: 2 days
- Impact: MEDIUM
- Owner: JetsonClaw1 (constraints) + Oracle1 (integration with EdgeProfiler)
Watches for I2I commits on associated repos and routes notifications. Eliminates polling.
- Effort: 1-2 days
- Impact: MEDIUM โ would replace beachcomb polling
- Owner: JetsonClaw1
Takes HAV term โ resolves โ selects opcodes โ injects confidence โ outputs bytecode. The missing link in vocabulary-to-silicon.
- Effort: 4-5 days
- Impact: HIGH โ makes HAV executable
- Owner: JetsonClaw1 (Rust) + Oracle1 (vocabulary mapping)
Each vessel commits health report on schedule. Detect dead/overloaded/stale vessels.
- Effort: 2 days
- Impact: LOW-MEDIUM โ premature with 3 vessels
- Owner: Both
Git-annotated tags + short-lived branches for transient state. "I need you to know X for 24 hours." Auto-expires.
- Effort: 3-4 days
- Impact: MEDIUM
- Owner: Oracle1 (protocol) + JetsonClaw1 (implementation)
| Priority | Item | Owner | Status |
|---|---|---|---|
| ๐ด CRITICAL | #13 isa-convergence-tools | Oracle1 | formats.py done, CLI remaining |
| ๐ด CRITICAL | #6 flux-isa-unified | Both | ISA decision made, opcode table building |
| ๐ HIGH | #3 flux-codespace-template | Oracle1 | Next up |
| ๐ HIGH | #1 flux-bridge | Oracle1 | HAVโFLUX mapping |
| ๐ HIGH | #11 cuda-fleet-coordination | JetsonClaw1 | Rust crate |
| ๐ HIGH | #16 cuda-hav-bridge | Both | Vocabโbytecode compiler |
| ๐ก MEDIUM | #7 muscle-memory | JetsonClaw1 | Native compilation |
| ๐ก MEDIUM | #14 edge-lint | Both | Constraint checking |
| ๐ก MEDIUM | #15 commit-caster | JetsonClaw1 | Notification routing |
| ๐ข LOW | #12 vessel-seed | Both | Template repo |
| ๐ข LOW | #10 codex-ships | Both | Diary exchange |