You ask your agent to write a comparison doc and it comes back with an executive summary, a detailed breakdown, a summary table that repeats everything in the breakdown, and a conclusion that restates the executive summary. Four representations of the same information, and half of it is filler like "it's worth noting" and "let's now explore."

Lean Exec teaches your agent the difference between redundancy and specificity. Redundancy is waste: saying the same thing twice in different formats, narrating what a table already shows, restating the user's question back to them. Specificity is value: concrete numbers instead of "very fast," edge cases that save debugging time, "when NOT to use this" guidance, rationale for decisions.

The skill defines 12 specific patterns to cut. Duplicate representations (table AND prose covering the same points). Bookend sections (executive summary plus conclusion saying the same thing). Meta commentary ("in this document we will explore"). Unrequested auxiliary files. Scope creep sections the user didn't ask for. Pre action announcements ("I'll now read the file"). Hedge stacking. Conversational padding after generating files.

It also defines what to always keep: specific numbers over vague qualifiers, edge cases and failure modes, nuance and caveats, actionable next steps. The goal is never to compress away the things that make output genuinely useful.

The skill scales to complexity. Quick answers get a few sentences. Deep analysis gets full depth with all relevant specifics. But neither gets redundancy or filler.

Does not apply to code output. Error handling, type hints, test coverage, and comments that explain why are never cut.

Works with Claude Code, Cursor, Codex CLI, VS Code Copilot, and any agent that supports the SKILL.md standard.