KetraTermModule ketraterm-completion
This module owns command-line tokenization, semantic context resolution, structured source evaluation, deterministic ranking, and bounded in-memory learning. It depends on Kotlin coroutines and has no terminal-module dependencies; source evaluation does not require a terminal session or UI. The README supplies the consumer entry point.
Source layout
| Package | Responsibility |
|---|---|
io.github.ketraterm.completion.api |
Engine and source contracts, source factories, requests, candidates, shell capabilities, messages, and the learning store. |
io.github.ketraterm.completion.model |
Command specifications, dynamic value domains, and immutable learning data exchanged with hosts. |
| Implementation packages | Tokenization, context resolution, matching, indexing, candidate projection, and ranking. Their declarations are internal. |
External consumers use api and model. The engine resolves one semantic
context per request and shares it across sources; providers must not reconstruct
that context with their own tokenizer. Its command catalog also supplies the
static source, keeping parsing and static suggestions consistent. Custom engines
use the public TerminalCompletionContext.resolve operation with their own
catalog and retain the resulting request-owned context across suspension.
Integration boundaries
Hosts supply immutable request metadata, bounded suspending providers, command outcomes, and explicit feedback. They own I/O dispatching, request cancellation, editing admission, presentation, and persistence policy.
Related modules own the adapters:
ketraterm-completion-host: path resolution and bounded host filesystem access.ketraterm-ui-swing-host: completion request/candidate and feedback conversion.ketraterm-completion-persistence: product storage and write coordination.
The detailed privacy, context, evaluation, and ranking contracts live in the completion architecture guide.