name	anti-hall-simplify
description	Behavior-preserving simplification workflow for Codex. Use when the user asks to simplify, deslop, trim fat, or reduce over-engineering.

anti-hall simplify for Codex

Scope the simplification to the named files or current diff. Do not sweep the whole repo unless explicitly asked.

Workflow:

Establish a green baseline by running the relevant check.
List simplification findings using tags:
- delete: dead or unreachable code
- stdlib: hand-rolled standard library behavior
- native: language builtin/idiom replacement
- yagni: speculative abstraction with no current caller
- shrink: verbose equivalent that can be clearer shorter
- slop: redundant comments, ceremony, impossible defensive checks
Apply only behavior-preserving edits.
Re-run the same check.
Report measured net: -N lines from actual diff output, never projected savings.

Model routing:

audit/selection: gpt-5.6-sol when ambiguous or risky
implementation: gpt-5.6-terra
simple mechanical cleanup: gpt-5.4-mini (default; gpt-5.6-luna available when 5.6-era capability/cutoff matters) — gpt-5.3-codex-spark is a distinct, faster/less-capable model, ChatGPT Pro only

If behavior would change, decline that finding and say why.

name	anti-hall-simplify
description	Behavior-preserving simplification workflow for Codex. Use when the user asks to simplify, deslop, trim fat, or reduce over-engineering.

Scope the simplification to the named files or current diff. Do not sweep the whole repo unless explicitly asked.

Workflow:

Establish a green baseline by running the relevant check.
List simplification findings using tags:
- delete: dead or unreachable code
- stdlib: hand-rolled standard library behavior
- native: language builtin/idiom replacement
- yagni: speculative abstraction with no current caller
- shrink: verbose equivalent that can be clearer shorter
- slop: redundant comments, ceremony, impossible defensive checks
Apply only behavior-preserving edits.
Re-run the same check.
Report measured net: -N lines from actual diff output, never projected savings.

Model routing:

audit/selection: gpt-5.6-sol when ambiguous or risky
implementation: gpt-5.6-terra
simple mechanical cleanup: gpt-5.4-mini (default; gpt-5.6-luna available when 5.6-era capability/cutoff matters) — gpt-5.3-codex-spark is a distinct, faster/less-capable model, ChatGPT Pro only

If behavior would change, decline that finding and say why.

anti-hall-simplify