Skip to content

Graphenium v0.6.0: Token optimization, composite tools, trust gating

Choose a tag to compare

@lambda-alpha-labs lambda-alpha-labs released this 27 Jun 12:06

Graphenium v0.6.0

Token optimization, composite MCP tools, and stricter trust gating.

v0.6.0 focuses on reducing context-window waste, adding higher-order query tools, and giving agents stricter trust controls.

New composite MCP tools

  • analyse_symbol(symbol) — Single-turn symbol analysis: node metadata + behavioral connections (calls, uses, inherits) + structural (imports, contains) + trust profile
  • module_dependencies(a, b) — Module-to-module dependency summary between two directory paths
  • what_changed(snapshot) — Risk-sorted delta against a stored snapshot: removed symbols, community moves, additions, downstream impact

Token optimization

  • summarize_file(show_leaves=false) — Hubs (degree > 5) shown by default; low-degree leaves hidden with expand instruction. Payload drops from 30KB to under 3KB for large files.
  • Trust Profile appended to every query response: N EXTRACTED, N INFERRED, N AMBIGUOUS

Stricter trust gating

  • include_tests=false (default) — Test/spec nodes excluded by default across all query tools
  • extracted_only=true on get_neighbors — Strict mode with zero heuristic/ambiguous edges

Documentation

  • README and SKILL.md both updated with new tool tables and parameter descriptions