aiken-verify

Aiken verify-before-trust

Aiken is a strict, statically-typed language for Cardano validators. A snippet that looks right is not the same as a snippet that compiles — indentation, type signatures, and stdlib imports in Aiken are all points where confident, plausible-looking code silently fails to build. Never present Aiken code as correct without actually compiling it.

Workflow

1. Write or gather the full Aiken snippet (a validator, a helper function, a test) that is about to be shown to the user.
2. Call validate_aiken_snippet (from the cardano-mcp server bundled with this plugin) with that exact snippet. It shells out to the real aiken build compiler in a temporary project and returns actual compiler output — not a guess.
3. Check the returned status field — it's one of three states, not two:
   • "ok" — compiled cleanly. Say so plainly ("compiles cleanly against aiken build") before presenting the code.
   • "error" — a real compile failure. Read the compiler's error output, fix the code, and validate again before showing anything to the user. Do not show a failing snippet as if it were final.
   • "unavailable" — the aiken CLI itself wasn't found in this environment; the code was not checked at all. Do not treat this the same as a passing or failing compile — say explicitly that verification couldn't run and the code is unverified, then fall back to your best understanding of Aiken syntax, clearly flagged as such.
4. If the cardano-mcp MCP server isn't available in this session at all (the tool call fails outright rather than returning an unavailable status), say so explicitly and fall back the same way. Never imply a snippet was compiled when it wasn't.

Why this exists

Generic LLM-written Aiken is frequently wrong in small, confident-looking ways — a missing use aiken/builtin, a validator returning Bool where a True/False-shaped constructor is expected, or a datum shape mismatch that only shows up as a compile error. A verification step against the real compiler beats a model's confidence about whether code is correct, every time.