Skip to content

Review and publish

Agent Lexicon separates discovery from governance. A scan can produce candidates and evidence, but only reviewed decisions become published vocabulary.

alex init creates a SQLite-backed workspace under .agent-lexicon/. The workspace stores:

  • scan results and line-numbered evidence;
  • candidate status;
  • reviewer decisions;
  • provenance records;
  • publication metadata.

The canonical lexicon remains a normal git-tracked file under lexicon/.

Terminal window
alex review

Use the review UI to distinguish:

  • a real project concept that should become canonical;
  • an alias of an existing term;
  • a deprecated surface that agents should avoid;
  • example-local or generated language that should be rejected;
  • an unresolved candidate that needs more evidence.
Terminal window
alex publish

Runtime decisions include a content-addressed snapshot reference. That makes a decision replayable against the same vocabulary content later.

Terminal window
alex workspace export-decision-log \
--root . \
--action review_decision_saved

The export is useful for audits, handoff, or external analysis. It is not a substitute for the canonical lexicon file.

Optional semantic components may suggest a likely canonical neighbour. They never silently accept or publish a term. The reviewer remains the authority for vocabulary state.