Deslop Agent Skill

Remove AI Code Slop

Check the diff against main and remove AI-generated slop introduced in the branch.

Trigger

Use after completing changes, before committing, or when code feels over-engineered.

Commands

git fetch origin main
git diff origin/main...HEAD --stat
git diff origin/main...HEAD

Workflow

• Run diff commands to see all changes on the branch.
• Identify slop patterns from the focus areas below.
• Apply minimal, focused edits to remove slop.
• Re-run git diff origin/main...HEAD to verify only slop was removed.
• Run tests or type-check to confirm behaviour unchanged: npm test -- --changed --passWithNoTests 2>&1 | tail -10
• Summarise what was cleaned.

Focus Areas

• Extra comments that state the obvious or are inconsistent with local style
• Defensive try/catch blocks that are abnormal for trusted internal code paths
• Casts to any used only to bypass type issues
• Over-engineered abstractions for one-time operations (premature helpers, factories)
• Deeply nested code that should be simplified with early returns
• Backwards-compatibility hacks (renamed _vars, re-exports, // removed comments)
• Features, refactoring, or "improvements" beyond what was requested
• Added docstrings, type annotations, or comments on code that wasn't changed
• Error handling for scenarios that can't happen in trusted internal paths

Guardrails

• Keep behavior unchanged unless fixing a clear bug.
• Prefer minimal, focused edits over broad rewrites.
• Three similar lines of code is better than a premature abstraction.
• If you remove something, verify it's truly unused first.
• Keep the final summary concise (1-3 sentences).

Skill-file mode

When the target is a SKILL.md (not a code diff), lint it against the same

slop instinct applied to prose. Run this before promoting a skill. Flag:

• Stale lines - guidance written for an old version of the skill that no

longer matches what it does. Cut it.

• Bloat - the skill runs past one screen with detail that belongs in a

linked reference. Push it down: in-skill step, then in-skill reference, then

an external file behind a pointer.

• Dead sentences - a line that changes nothing if deleted. Delete it.
• Duplication - the same instruction stated in two places, so edits drift.

Keep one source of truth.

• Premature stop - the method ends before the work does (asks the question

but never records the answer, cleans but never verifies).

• Weak anchor - no single concept the skill turns on. A skill the reader

can name in one word triggers and executes in fewer tokens.

• Missing invocation intent - no declared human-run vs auto-triggered mode.

See rules/skill-conventions.mdc.

• Wrong write op - a state-changing skill that does not say whether it

adds, updates, or appends, or that duplicates its output on a second run.

Skill-file mode output: the flagged issues by line, the edits applied,

the cleanup summary, and whether the skill is ready to promote.

Output (code mode)

• List of slop patterns found with file locations
• Edits applied
• One-line summary of what was cleaned