Inspiration
Inscription minting on Solana is fragmented: a creator who wants to launch an inscription collection must wire together the SPL-20 JSON convention, the on-chain write, an NFT carrier, a mint page, and (if they want trading) a marketplace, each from scratch. There is no small, honest, reusable starting point — and the existing inscription tooling is hard for a newcomer to try without spending real money. Skelsol takes its shape from Spooky B-Agents, a BSC inscription game played by AI agents: a mint page with live stats, a leaderboard, and a market — plus a CLI so agents can play without a browser. Skelsol rebuilds that shape on Solana as a reusable template, with all-new code and all-new art, and with the honesty the original never needed to write down: what is an on-chain guarantee, and what is a convention.
What it does
Skelsol is one lean template that a creator configures (ticker, supply, price, image) and gets:
- A mint client that writes an SPL-20-style inscription — a JSON record in the {"p":"spl-20","op":"mint","tick":"...","amt":"..."} shape — into the memo of a Solana transaction, and mints a Metaplex NFT as the inscription's carrier. One mint, one inscription, one carrier NFT.
- A reader that rebuilds collection state (minted, minters, leaderboard, live activity) from on-chain records.
- Two working web applications built from the template: Ghost Mint (mint ghost inscriptions, live stats, wallet lookup, minter leaderboard, live activity) and the Inscription Market (list, buy, and cancel inscription sales at fixed prices, with the NFT held in on-chain escrow until it is bought or cancelled).
- A CLI (skel) so AI agents can use the same template from a terminal: status, mint, list, buy, cancel, lookup, with --json output and a local keypair wallet. A manual for agents (agent.md) ships at the repo root.
- A minimal Anchor escrow program (skelsol-market) so the market trades without trust: buy pays 99% to the seller and 1% to the treasury in one atomic transaction; only the seller can cancel. The demo collection is GHOST ("Spooky Ghosts"): supply 4,444, 4 inscriptions per wallet, free mint (rent deposits only). Both demo apps and the CLI default to devnet, so anyone can try the whole flow for free; a network switch moves each app to mainnet.
How it was built
- TypeScript everywhere. The template library is UI-free so the web apps and the CLI share one implementation. Tests run with vitest; the apps compile with tsc to static sites (no server, no database — all state is on-chain).
- Inscriptions via the Memo program. The JSON is written with a Memo v3 instruction in the mint transaction; the carrier is a Metaplex Token Metadata NFT minted with the Umi SDK. No custom mint program — minting needs no deploy and no audit. The choice of Token Metadata over Core is deliberate (every wallet, explorer, and marketplace reads it) and recorded in ADR 0002.
- Escrow with Anchor 1.2.2. The market program follows the canonical make/take/refund escrow pattern, adapted to a fixed-price NFT-for-SOL sale with a 1% fee. Its TypeScript client builds the list / buy / cancel instructions by hand from the committed IDL, because there is no Anchor 1.x TypeScript client on npm (ADR 0010). The program is tested with 8 in-process LiteSVM tests, all passing.
- Wallet Standard for browser wallet connection (Phantom, Solflare), with legacy/v0 transactions so Ledger signing keeps working.
- Design from primary-source research. Before building, the project researched LibrePlex SPL-20, the Memo program, Metaplex Token Metadata, the Anchor escrow pattern, and devnet in the official docs and source code (docs/research-solana-inscriptions.md). That research is why the project says "SPL-20-style" and documents its limits honestly instead of claiming to implement LibrePlex SPL-20.
- Spec-driven, in Kiro. Requirements, design, and tasks live in .kiro/specs/skelsol/; decisions live in docs/adr/ (10 ADRs).
Challenges
- There is no SPL-20 spec. Research found no published specification from LibrePlex, no maintained reference in 2026, and that real LibrePlex inscriptions live in program-owned accounts, not in memos. The honest answer was to label the project "SPL-20-style", document the memo-NFT binding as a convention the chain does not enforce, and say plainly that no existing SPL-20 indexer will recognize these inscriptions.
- The devnet faucet is closed to scripts. solana airdrop against the public RPC fails, and the public RPC faucet returns "closed" — so funding (for the deploy wallet and for judges) goes through the web faucet instead, and the docs say so.
- No Anchor 1.x TypeScript client. The escrow program is built with Anchor 1.2.2, but npm has no matching TypeScript client, so the market client encodes instructions and decodes Listing accounts by hand, with tests pinning the discriminators to the IDL.
- Rent is a moving target. SIMD-0437 is cutting rent in steps through late 2026, so no rent figure is hard-coded anywhere; the mint client queries the live rent minimum at runtime.
- No browser wallet in the build environment. The hosted apps have been checked in a real browser (stats, market status, listings — the market was verified live on devnet on 19 Nov 2026), but the wallet handshake and a live mint with Phantom/Solflare still need a human with a wallet in a real browser; the README says so.
What's next
- Host real collection art (the config currently points at a placeholder image URL) and the off-chain metadata JSON on durable storage (Arweave/IPFS).
- A real-browser test pass with Phantom and Solflare: the wallet handshake and a live mint are the only parts of the demo flow not yet exercised in a real browser (the market program is deployed, both apps are hosted, and the stats and market status were re-checked on 19 Nov 2026).
- Metaplex Core migration path for lower mint cost if the compatibility trade-off ever flips (documented in ADR 0002).
- On-chain inscription accounts (LibrePlex or Metaplex mpl-inscription) if a future feature ever needs a program to read inscription content — memo data cannot be read on-chain.
Honest limits
- "SPL-20-style", not LibrePlex SPL-20. The inscription JSON is written into a transaction memo; real LibrePlex inscriptions live in program-owned accounts. No existing SPL-20 indexer or marketplace will recognize these inscriptions, and there is no published SPL-20 specification.
- Memo inscriptions live in transaction logs, not account state. Verifiable from the transaction signature, but no on-chain program can read them.
- The memo-NFT binding is a documented convention (same transaction, plus the mint address in the JSON's mint field), not something the chain enforces.
- Supply caps and per-wallet limits are client-side rules, enforced by the reader counting on-chain inscriptions. There is no custom mint program.
Built With
- anchor
- kiro
- metaplex
- node.js
- solana
- typescript
- vitest
