topos.mcp.tools.assess¶

Assessment tool — compare current vs. proposed code on the lattice.

This is the main tool for agent refactor loops. When filepath is provided, the baseline is evaluated against the cached ModuleDependencyGraph and the proposed AST is scored against that same graph (approximating coupling under the refactor). Anti-gaming guardrail: if scores moved meaningfully while AST edit distance is near zero, status becomes SUSPICIOUS_NO_STRUCTURAL_CHANGE.

topos.mcp.tools.assess.topos_assess_improvement(params)[source]

Compare proposed code against the current baseline.

Preferred usage — pass filepath (code loaded from disk + coupling scored against the cached ModuleDependencyGraph). The proposed code is parsed, but coupling is an approximation: it uses the current dep graph for the target file, so inbound edges from other files reflect the pre-refactor state. That’s fine for tight iteration loops.

Legacy usage — pass current_code + proposed_code. Coupling is NOT computed (AST-only).

Anti-gaming: when scores move meaningfully but AST edit distance is near zero, status becomes SUSPICIOUS_NO_STRUCTURAL_CHANGE and suspicion_reason is populated.

topos.mcp.tools.assess.render_assessment_md(r)[source]

Compact markdown for a refactor assessment.

Summarizes current vs. proposed rather than dumping both full evaluations; the structured_content channel still carries everything.