Shuriken packages LLM-consumable skill files (SKILL.md, Claude Code format) and a Rust crate so coding agents can authenticate, scope permissions, and call a multi-asset trading API.
Key Takeaways
Skills surface as shuriken:<skill-name> and install via native plugin commands for Claude Code, Codex CLI, Copilot CLI, Gemini CLI, Cursor, and OpenCode.
The Rust crate (shuriken-skills) bakes skill markdown into a binary at compile time; shuriken-api consumes it internally via the same git dependency.
Granular agent-key scoping is central to the design: no seed phrases in the agent runtime, least-privilege enforced via shuriken:scoping skill.
v0.1 covers three skills: api-integration, agent-keys, and scoping; release versioning must stay in lockstep across Cargo.toml, package.json, and all plugin manifests.
Supported asset classes include onchain tokens, perpetuals, real-world assets, pre-IPO equity, and prediction markets with data feeds from Twitter, Telegram, Discord, and on-chain activity.
Hacker News Comment Review
Discussion is minimal and jokey; no technical critique of the permission model, Rust crate design, or agent-key scoping approach has surfaced yet.
The humor centers on agents compounding financial risk on top of AI unpredictability, with the founders acknowledging the framing directly.
