arena-bridge
Skill8 starsChanged 3 months agoMITView on GitHub
Skill and protocol specification for AI agents (Arena.ai Agent Mode, Claude Code, Codex, Cursor, and custom agentic frameworks) interacting with Skainet Bridge. Skainet Bridge is an authenticated, local-first bridge that breaks AI agents out of restrictive sandboxes (such as Arena.ai's 128 MB snapshot cap and GPU/X11 absence) and connects them directly to the operator's real machine, local tools, file system, and hardware. This skill equips agents with the exact operational protocol: 1. Connecting to the bridge (Local loopback, Tailscale Funnel,…
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- book-of-eternitySkill
- browseractSkill
- build-capabilitySkill
- cleanupSkill
- digestSkill
- healthSkill
- snapshotSkill
- auto-fixSkill
- sys-snapshotSkill
- researchSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 33
Every request requires the master token in the Authorization header:
- Very specific: line 69
Host Workspace Path: Default host storage is located under ~/arena-bridge/ on the operator's PC.
- Very specific: line 83
When building games with Godot (e.g.
- Distinctive: line 96
When acting as an autonomous maintainer or feature builder:
- Very specific: line 98
Task Board First: Every task is assigned an identifier from T0 to Tn in docs/TASK_BOARD.md.
- Very specific: line 100
Root-Cause Fix: Never silence errors with || true or blanket suppressions.
- Very specific: line 106
Use Serena MCP (serena start-mcp-server) or session recall (arena/agentctl_cli/agentctl_memory.py) to persist architectural decisions and …
- Very specific: line 108
Read AGENTS.md and docs/TASK_BOARD.md at the beginning of every turn.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
arena-bridge (Skainet Bridge Skill)
Skill and protocol specification for AI agents (Arena.ai Agent Mode, Claude Code, Codex, Cursor, and custom agentic frameworks) interacting with Skainet Bridge.
1. Overview & Purpose
Skainet Bridge is an authenticated, local-first bridge that breaks AI agents out of restrictive sandboxes (such as Arena.ai's 128 MB snapshot cap and GPU/X11 absence) and connects them directly to the operator's real machine, local tools, file system, and hardware.
This skill equips agents with the exact operational protocol:
- Connecting to the bridge (Local loopback, Tailscale Funnel, Cloudflare, ngrok, bore);
- Bypassing sandbox filesystem limits (persistent storage on host, >128 MB);
- Offloading heavy compute, game engines (e.g. Godot), and GPU tasks to the host;
- Executing MCP tools and managing long-running background processes;
- Structured task execution via Spec-Kit (T0..Tn) and Serena session recall.
2. Discovery & Connection
Connection Transports
The bridge exposes an HTTP/REST API and an MCP (Model Context Protocol) endpoint:
- Local host:
http://127.0.0.1:8765 - Tailscale Funnel:
https://<node-name>.ts.net - Public Tunnels: ngrok (
https://<id>.ngrok-free.dev), Cloudflare Quick Tunnels, or bore.
Authentication
Every request requires the master token in the Authorization header:
Authorization: Bearer <TOKEN>Tip: On first launch, start.bat / start.sh automatically generates token.txt
and copies the token to the clipboard.
Diagnostics & Handshake
To verify connectivity from inside the agent sandbox:
python scripts/check_bridge.pyOr query endpoints directly:
# 1. Check health & version
curl -s http://127.0.0.1:8765/health
# 2. Authenticated self-descriptor (profile & active tools)
curl -s -H "Authorization: Bearer $(cat token.txt)" http://127.0.0.1:8765/v1/self
# 3. Active tunnel & external URL status
curl -s -H "Authorization: Bearer $(cat token.txt)" http://127.0.0.1:8765/v1/tunnels/status3. Sandbox Escape & Large Storage Protocol
In sandboxed environments (like Arena.ai Agent Mode), the root directory /home/user
is strictly capped at 128 MB and 10,000 files between messages. Exceeding this budget
causes catastrophic session truncation.
The Bridge Storage Solution:
- Workspace Isolation: Temporary files during the session are written to
/tmp(1 GB tmpfs). - Persistent Artifacts on Host: Heavy files (large datasets, media assets, Godot game binaries, APKs)
are streamed to the host machine via Bridge
/v1/files/*or MCP filesystem tools. - Host Workspace Path: Default host storage is located under
~/arena-bridge/on the operator's PC.
+---------------------------+ +-------------------------------+
| Arena.ai Sandbox | | Operator Host Machine |
| (128 MB ephemeral snapshot| HTTP / MCP | (Unlimited disk, real GPU, |
| /tmp for tooling) | ----------> | Godot engine, hardware) |
+---------------------------+ +-------------------------------+4. Godot Game Production via Bridge
When building games with Godot (e.g. using godot-game-production-skill):
- Engine Execution: Godot runs on the host machine where real display/GPU or headless binaries reside.
- Headless Harness: The agent sends GDScript files to the host via Bridge, runs headless unit/smoke tests, and collects structured game state.
- Visual Frame Verification: Visual frames and game screenshots are captured via Bridge desktop/window
probes (
/v1/desktop/screenshotor/v1/desktop/window/*) and returned to the agent. - Asset Management: 3D models (.gltf), sound banks, and texture packs live entirely on the host
filesystem under
~/arena-bridge/projects/<game_name>/, never bloating the 128 MB sandbox.
5. Agent Workflow & Discipline (Spec-Kit & DoD)
When acting as an autonomous maintainer or feature builder:
- Task Board First: Every task is assigned an identifier from
T0toTnindocs/TASK_BOARD.md. - Strict Definition of Done (DoD):
- Root-Cause Fix: Never silence errors with
|| trueor blanket suppressions. - Isolated Parity Suite: Deterministic unit/integration tests with mocks for external processes and sockets.
- Zero Mutation Survivors: Tested against
mutmutwith 0 surviving mutants. - Mandatory Bilateral Sabotage: Intentionally breaking the fix fails tests; healthy code is 100% green.
- Preflight Pass:
python scripts/preflight.py(21 local gates) must be completely green before push.
- Root-Cause Fix: Never silence errors with
- Serena Session Memory:
- Use Serena MCP (
serena start-mcp-server) or session recall (arena/agentctl_cli/agentctl_memory.py) to persist architectural decisions and context across session boundaries. - Read
AGENTS.mdanddocs/TASK_BOARD.mdat the beginning of every turn.
- Use Serena MCP (
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

