AgentSkillsCN

negentropy

建立系统化的代码清理与整合常规流程

SKILL.md
--- frontmatter
name: negentropy
description: Systematic code cleanup and consolidation routine
user-invocable: true
argument-hint: [scope]

Announce: 🧹 NEGENTROPY

Reduce disorder in the codebase. The optional argument narrows scope — a file, directory, or module name. Without an argument, scope to files touched in this session.

Read every file in scope before proposing any changes.

Pass 1 — Compression

For each function, type, test, and module in scope, ask:

  • •Can this be deleted without changing behavior?
  • •Can this merge into an existing neighbor?
  • •Does duplication here reach three occurrences? If so, consolidate.

Propose deletions and merges. Block for confirmation before applying.

Pass 2 — Decomposition

For each function in scope:

  • •Over 20 lines? Candidate for split — but only where semantics support a seam.
  • •Over 1000 lines in the file? Candidate for module extraction.
  • •Multiple jobs in one name? Candidate for separation.

Balance: a ~400-line file with clear internal structure beats six ~70-line files with re-export boilerplate. Ergonomics is the tie-breaker.

Propose splits and extractions. Block for confirmation before applying.

Pass 3 — Comments

Audit every comment in scope:

  • •Restates the code → delete the comment.
  • •Contains a code example → delete the example.
  • •Explains what or how → rewrite to explain why, or delete.
  • •Uses temporal language ("recently", "moved", "now") → rewrite as evergreen.
  • •Uses relative quality language ("new", "enhanced", "improved") → rewrite as evergreen.

Apply comment changes without blocking.

Pass 4 — Names

Read function and variable names as prose. Where the composed story is unclear, rename for literate clarity. Nudge surrounding code style toward project standards.

Propose renames. Block for confirmation before applying.

After all passes complete, summarize what changed and resume the prior task.