caveman-manage
An agent skill by JuliusBrussee, from JuliusBrussee/caveman. Tags: automation, governance, risk, workflow.
What it does
Inspect Caveman Cloud's experiment lifecycle and block unsafe execution. Use when asked to start, approve, cancel, promote or roll back a Caveman experiment.
Install
With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:
npx skills add JuliusBrussee/caveman --skill caveman-manageOr 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-manage ~/.claude/skills/caveman-manageSafety box score
Grade B, low risk: 11 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 is a read-only operator for Caveman Cloud experiments: it inspects lifecycle state and results, then recommends one action while explicitly blocking execution. The main risks are remote network access to the Caveman service and exposure to whatever experiment data the logged-in identity can read; it does not ship code or enable writes.
| Category | Level | Why |
|---|---|---|
| Code & shell execution | 1 (low) | It can invoke read-only CLI commands such as `caveman cloud experiments list/show/results`, but the skill explicitly says to "Never execute a lifecycle mutation" and only report recommendations. |
| Network access | 1 (low) | It queries Caveman Cloud over MCP or CLI (`caveman_context`, `caveman_experiment_get`, `caveman cloud experiments ...`), so it sends requests to a remote service, but only for reads. |
| Filesystem reach | 0 (none) | The skill only shows remote-read commands and does not read, write, or delete local files outside the working directory. |
| Secrets & credentials | 0 (none) | It does not ask for passwords, tokens, `.env` files, or other credentials; it relies on the logged-in Caveman identity and RBAC. |
| Privilege & persistence | 0 (none) | No sudo, global install, service registration, shell startup edits, or OS-level changes are described. |
| Instruction hijack surface | 0 (none) | The text constrains behavior rather than trying to override higher-priority instructions; it does not instruct the agent to follow external prompts or hide anything. |
| Irreversible actions | 1 (low) | It can recommend `start`, `approve`, `cancel`, or `rollback`, but it explicitly forbids actually executing lifecycle mutations, so autonomy is limited to advice. |
| Transparency | 0 (none) | The skill is plain markdown with no binaries, generated code, base64 blobs, or unpinned installers. |
Before installing
- Read `SKILL.md` to confirm it only uses read-only Caveman MCP/CLI calls.
- Test it with a non-sensitive Caveman account first to see what experiment data becomes visible.
- Verify the environment does not alias `caveman` to any wrapper that could do more than `list/show/results`.
- Confirm you are comfortable with the agent seeing experiment metadata returned by your logged-in Caveman identity.
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-manage/SKILL.md
- Branch
- main
- Updated
- 2026-10-05
Related skills
- automate-me — Use for "automate me", "create/update/refresh my -mode skill", "turn/capture my preferences or working style into a skill".
- cancel-ralph — Cancel an active Ralph Loop. Use when the user wants to stop, cancel, or abort a running ralph loop.
- churn-prevention — When the user wants to reduce churn, build cancellation flows, set up save offers, recover failed payments, or implement retention strategies.
- cli-for-agents — Designs or reviews CLIs so coding agents can run them reliably: non-interactive flags, layered --help with examples, stdin/pipelines, fast actionable errors.
- continual-learning — Orchestrate continual learning by delegating transcript mining and AGENTS.md updates to `agents-memory-updater`.
- create-verification-skill — Generate a project-local verification skill that drives your app the way a user does — any language, framework, or platform.