complexity-mitigator

name

complexity-mitigator

description

Existing-code comprehension and local winnowing preflight. Use for simplify/refactor/clean up/untangle, nested branches, boolean soup, opaque names, mixed responsibilities, cross-file state, or review stalls. Factor the local whole, separate essential/incidental/specification-risk factors, winnow dominated or duplicated factors, and emit the smallest clarity cut. Not for broad architectural layer removal, kernel quotienting, invariant remediation, or greenfield planning.

Complexity Mitigator

Intent

Reduce understanding cost in existing code while preserving essential domain meaning. This skill is a read-only routing rail and analysis preflight, not a delivery owner.

Doctrine

Use a lightweight local form of:

FACTORING -> WINNOWING

• Factor the code into responsibilities, states, effects, decisions, and boundaries.
• Retain essential domain factors.
• Expose specification-risk factors before refactoring.
• Remove, collapse, rename, localize, or delegate incidental factors.
• Do not run full behavioral quotienting unless repeated distinctions and accepted observations make it necessary; route that to review-fold or the owning review-closure workflow.

Activation

Use for:

• hard-to-follow existing code;
• deep nesting, boolean soup, flags, mixed parse/validate/decide/effect;
• cross-file hops, hidden state, ordering dependencies, opaque names;
• review stalls caused by comprehension;
• explicit local simplicity/refactor requests.

Handoff instead:

• broad layer/framework/tooling tax -> reduce;
• unclassified review distinctions -> review-fold;
• classified owner-boundary pressure -> $actuating after $review-fold authors the current Counterexample Set;
• illegal states / invariant ownership -> invariant-ace;
• missing essential structural shape -> universalist;
• implementation -> $actuating or the owning workflow.

Modes

Micro Preflight

Use when another workflow owns implementation.

Complexity preflight:
- whole:
- dominant cost:
- factorization:
- winnow:
- smallest clarity cut:
- do not change yet:
- handoff:

Compact

For one function, file, or review comment:

1. Slice
2. Heat Read
3. Factorization / Winnowing Receipt
4. Essential vs Incidental vs Specification Risk
5. Preferred cut
6. Proof signal
7. Handoff

Full

Use only when the slice crosses files, behavior is unclear, specification risk is material, or the user asks for deep analysis.

Definitions

• Essential complexity: domain rules, invariants, required state transitions, and irreducible external obligations.
• Incidental complexity: complexity introduced by implementation choices; reducing it preserves required behavior.
• Specification risk: uncertainty about required behavior, policy, or boundary contract.

Workflow

1. Choose the slice: entrypoint, inputs, outputs, state, effects.
2. Heat-read nesting, branches, flags, cross-file hops, state to simulate, and hidden side effects.
3. Factor the whole into responsibilities, decisions, state, effects, external constraints, and proof surfaces.
4. Classify each factor as essential, incidental, mixed, or spec-risk.
5. Check whether incidental factors are duplicated, dominated, subsumed, vestigial, pass-through, or better delegated to an existing facility.
6. Preserve essential factors and unresolved specification-risk factors.
7. Winnow in this order: delegate, flatten, rename, localize, collapse, extract only after stable repetition appears.
8. State the recomposition rule: how the retained factors still explain the behavior.
9. Name the smallest proof signal and handoff.

Review finding handoff

When an actuation review profile selects this lens, return factorization evidence to the review caller. $review-fold may cite it when classifying a counterexample-set/v1; $actuating alone evaluates that Counterexample against the current Construction and selects any successor. Do not choose the repair, architecture, work node, or mutation.

complexity_counterexample_evidence:
  owner_boundary:
  governing_law:
  participating_abstractions:
    - abstraction:
      live_obligation:
      status: retain | retire | collapse | delegate | replace | validate-first
  dominated_factors: []
  smallest_local_repair:
  local_repair_adds_semantic_machinery: true | false
  structural_pressure: []
  proof_surface_before: []
  proof_surface_after: []
  falsifier:
  evidence_refs: []

If local repair adds a protocol, state, helper abstraction, repeated branch family, or wound-specific test family, mark it as semantically growing and record the structural pressure. Actuating decides whether the current Construction is a realization, architecture, or ablation defect.

Factorization / Winnowing Receipt

factorization_winnowing_receipt:
  whole: "..."
  factors:
    - id: "F1"
      obligation: "..."
      class: essential | incidental | mixed | spec-risk
      owner: "..."
      evidence_ref: "..."
      status: retain | delete | collapse | delegate | localize | rename | validate-first
  duplicated_or_dominated_factors: []
  irreducible_core: []
  recomposition_rule: "..."
  preservation_relation: observationally-equivalent | refinement-preserving | unknown
  proof_signal: "..."
  handoff: "..."

Guardrails

• No file edits or commits.
• Do not select a review repair or work node.
• Do not confuse lower LOC with lower understanding cost.
• Do not delete essential policy or unresolved external obligations.
• Do not merge accidental rhymes.
• Do not extract abstractions before stable shape is visible.
• If behavior is unclear, recommend a learning artifact: example matrix, contract test, fixture set, state table, or spike.
• If reduction crosses architectural layers, hand off to reduce.

Local playbook

1. Delegate homegrown incidental logic when a current local/platform facility has matching semantics.
2. Flatten control flow while keeping essential protocol/lifecycle steps explicit.
3. Rename vague verbs and implicit state.
4. Move meaning into data when it reduces branching or invalid combinations.
5. Extract only stable repeated concepts; one case is unique, two may be coincidence, three may justify extraction.

Output

End with:

Winnowing Bottom Line:
- retain:
- remove/collapse/delegate:
- unresolved specification risk:
- recomposition:
- proof:
- handoff: