Add a language feature

Begin with the source-level rule and its rejected boundary. Decide whether the change is syntax, semantics, library behavior, or only an optimization; each has a different earliest owner.

Write three examples before editing: the smallest accepted program, the closest rejected program, and one interaction with an existing feature. These become the positive test, negative diagnostic, and regression boundary. If the proposed behavior can be expressed entirely in std/*.psm, it is a library change and should not add a token or AST node.

Implementation path

1. Add tokens in src/lexer/token.psm and scanner behavior only when new vocabulary is required.
2. Add the smallest parser production and recovery boundary under src/parse.
3. Extend AST nodes or semantic types without storing backend-only LLVM details.
4. Resolve names, types, overloads, and flow in src/sema.
5. Define moves, borrows, drops, capture, container, and call behavior.
6. Teach AIF about new allocation sites, aliases, escapes, fields, or thread edges.
7. Lower the semantically resolved form under src/ir.
8. Extend the runtime or LLVM bridge only when ordinary Prismio or existing IR operations cannot express the capability.

Concrete entry points

New punctuation or keywords begin in src/lexer/token.psm; scan() in scanner.psm must produce the token with file, line, column, and length. Route the grammar through parseDeclaration(), parseStatement(), or Pratt-style parseExpression(). Use parserNodeFrom() to preserve the opening token's span and parser helpers for expected-token diagnostics and synchronization.

AST representation belongs in src/ast/nodes.psm and types.psm. Add serialization to src/ast/dump.psm when the node survives parsing: the AIF oracle consumes dump-ast, so an omitted field can make the independent model analyze a different program.

Semantic order matters. Named types are registered before function predeclaration; function bodies are checked by semaFunction(), expressions by semaExpr(), and statements by semaStatement(). Reuse semaTypesMatch(), semaExpectAssignable(), overload resolution, and flow helpers. Do not encode a source validity rule only in src/ir: prismio check deliberately stops before code generation.

Memory semantics need an explicit answer for every operand and result: copy, borrow, mutable borrow, consume, alias, or produce. Update the ownership visitor and AIF walk for allocations, points-to edges, returns, container retention, and thread transfer. Check the resulting site with prismio aif --manifest and --why, not merely the tier total.

Lower the resolved node with generateExpression() or generateStatement() and existing ir_* builder operations. Module-wide constructs also need declaration staging in generateModule() or generateFunction(). A new LLVM bridge function requires matching Prismio extern declaration, handle validation, C wrapper implementation, and bridge test.

Proof

Add focused positive and negative tests, multiple-error recovery where applicable, ownership and AIF cases, native execution, and artifact assertions for representation claims. Generic features need multiple specializations and a recursion/coherence boundary. Platform-facing features need the supported target matrix.

Run the test once with the last known-good generation and once with the candidate. If the compiler compiles itself differently, use the fixed-point gate to determine whether the difference stabilizes or keeps moving. A feature is not complete while dump-ast, check, native build, and the relevant analysis command disagree about it.

Update the language documentation and if user-visible and this developer portal for implementation behavior. Reserved syntax alone must never be labeled implemented.