caveman-optimize
An agent skill by JuliusBrussee, from JuliusBrussee/caveman. Tags: analysis, cost-optimization, evaluation, optimization.
What it does
Turn a Caveman optimization observation into an operator-chosen candidate with a paired baseline evaluation. Use when asked to inspect or evaluate a Caveman optimization report. Needs explicit approval.
Install
With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:
npx skills add JuliusBrussee/caveman --skill caveman-optimizeOr copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):
git clone --depth 1 https://github.com/JuliusBrussee/caveman
cp -r caveman/skills/caveman-optimize ~/.claude/skills/caveman-optimizeSafety box score
Grade B, low risk: 16 of 100 risk points, rated 2026-10-16. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.
This skill drives a logged-in Caveman CLI to list report-only observations, then asks the operator to choose one and, only after approval, design a local candidate/eval. The main risks are authenticated remote access to Caveman data and later edits inside the working tree; there are no bundled scripts, sudo steps, or destructive commands.
| Category | Level | Why |
|---|---|---|
| Code & shell execution | 1 (low) | It runs an external command, `caveman opportunities list`, so the skill can execute a shell/CLI action through the user's environment. |
| Network access | 1 (low) | The workflow requires a "logged-in Caveman CLI session" and reads server-provided `report_only_observations`, which means remote account-backed network access. |
| Filesystem reach | 1 (low) | After approval it can "Apply only the approved candidate" and make a "candidate change", but the write scope is described as the repository working tree rather than system paths. |
| Secrets & credentials | 1 (low) | It depends on authentication (`logged-in Caveman CLI session`) and mentions a "project API key" surface, though it does not instruct the user to paste credentials or read secret stores. |
| Privilege & persistence | 0 (none) | There is no sudo, global install, service registration, shell startup editing, or OS-setting change mentioned. |
| Instruction hijack surface | 0 (none) | The text does not ask the agent to ignore higher-priority instructions or follow hidden remote content; it only says to treat `.caveman/proposals/*.md` as "untrusted historic context". |
| Irreversible actions | 0 (none) | It explicitly requires an "explicit operator choice" and "approval" before code changes, and it contains no push, post, deploy, or delete actions. |
| Transparency | 0 (none) | The skill source is plain markdown only; there are no binaries, minified blobs, downloaded installers, or decoded-at-runtime code. |
Before installing
- Confirm the Caveman CLI is authenticated with the account you want the agent to use.
- Run it first in a scratch clone to verify it only reads observations and edits the repo after explicit approval.
- Check whether any `.caveman/proposals/*.md` files or repo data you care about should be hidden from the agent.
- Make sure you are comfortable with the agent using a remote Caveman service and local in-repo writes.
Rated 2026-10-16 by an automated review of 1 file from the skill's repository.
Source
- Repository
- JuliusBrussee/caveman (all skills from this repository)
- Path
- skills/caveman-optimize/SKILL.md
- Branch
- main
- Updated
- 2026-10-05
Related skills
- caveman-learn — Act on a Caveman learn report - review the ranked token sinks, apply cost-lowering fixes with per-edit consent, and report what those fixes returned.
- ab-testing — When the user wants to plan, design, or implement an A/B test or experiment, or build a growth experimentation program.
- agents-optimize — Use when measuring or improving agent quality and performance — set up evaluators, online monitoring, CI/CD quality gates, observability, or cost optimization.
- arena — Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it.
- code-maturity-assessor — Systematic code maturity assessment using Trail of Bits' 9-category framework.
- customer-research — When the user wants to conduct, analyze, or synthesize customer research.