
from cauldron13
CLI and workflow tools to prepare, upload, and run Frostbite on-chain AI models on Solana, including deterministic account lifecycle and on-chain inference test
Provides a hands-on workflow for preparing AI models for on-chain execution with Frostbite on Solana. The skill documents CLI commands and sequences to initialize projects, convert and pack weights, build guest binaries, create deterministic seeded accounts, upload weights, and run on-chain inference (devnet). It also describes a TUI mode (Wizard/Manual), lifecycle helpers for account management, and guidance for invocation tuning and safety.
Use this skill when you need to deploy or test a Frostbite-compatible model on Solana: setting up manifests, preparing weights, initializing seeded deterministic accounts, uploading binaries/weights, and running inference on devnet. It is also useful for developers debugging account lifecycle and upload/invoke flows.
Likely useful in developer-focused agents and CLI-capable assistants (Copilot/Codex/Cursor/Git-aware agents) that can run or explain shell workflows.
The skill is marketed as a Solana Frostbite ModelKit tool but the actual SKILL.md body is a generic Skill Hub Manager that downloads and installs arbitrary skills from a remote registry. The skill instructs the agent to immediately execute $ARGUMENTS without asking, creating a shell injection risk. The install flow fetches unverified remote content and writes it to disk with no integrity checks. Metadata/body mismatch is a red flag — likely a bait-and-switch or poorly maintained entry.
Metadata describes a Solana on-chain ML tool but the actual content is a skill package manager. The auto-execute-$ARGUMENTS pattern is dangerous. Not quite malicious (security 42, above 40 threshold) but concerning. Would not recommend installing or using this skill.