Caveman Optimize

This skill turns one of Caveman's report-only optimization observations into a candidate code change, and it refuses to treat the observation as a savings estimate. You pick the observation, it inspects the repo for a specific callsite that could produce the aggregate shape, and it proposes one minimal edit plus a paired baseline/candidate eval on identical fixed inputs. Nothing gets edited until you approve both the candidate and the eval design. Retired observation ids are explicitly ignored, and any dollar claim has to come from provider-complete accounting, not from local token counts. The honest take: the guardrails are strict to the point of being a little slow, and it needs a fixed fixture and quality check to work at all. If your repo lacks those, expect it to stop early and tell you what to build first.

skills.sh installs: 120.5K
GitHub stars: 118.7K
last updated: Oct 26, 2026

View on GitHub

Install

$ npx -y skills add juliusbrussee/caveman --skill caveman-optimize --agent claude-code

Installs into .claude/skills of the current project.

Files

Evaluate an optimization observation

Use Caveman's report-only observations as diagnostic input. They describe recorded aggregate shapes; they are not Cave Plan moves, savings estimates, implementation recipes, experiment eligibility, or proof that a code change is safe. Keep the workflow operator-chosen and evidence-first.

1. Read the exact observations

Require a logged-in Caveman CLI session and run:

caveman opportunities list

Read only the report_only_observations array. Do not select from the lifecycle data array. Preserve each server-provided title and observation verbatim. Handle these exact repository-profile ids:

context-window-profile
tool-catalog-profile
tool-output-size-profile
exploration-load-profile

SKILL.md on GitHub

Details

Repository: github.com/juliusbrussee/caveman
First seen: Oct 26, 2026