Guide · Agent skill
How to use the CAIO Review for AI Plans agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The caio-review skill makes your AI agent review any plan that involves AI the way a demanding Chief AI Officer would. It is aimed at founders, product leads and teams deciding whether an AI feature is ready to ship, whether to sign a long AI vendor contract, or whether a use case falls under the EU AI Act. After this guide you will know what the skill does, how to install it, how to trigger it with a plain request, and what its limits are.
CAIO Review for AI Plansalirezarezvani/claude-skills · MIT Download the skill Quick start
- 1Download the ZIP. Get the caio-review ZIP from this page and unzip it on your computer.
- 2Place the skill folder. Move the unzipped folder into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project.
- 3Confirm the plugin and Python. Make sure the c-level-agents plugin is installed and Python is available to run the referenced scripts.
- 4Prepare the JSON inputs. Create a use case file and a workload file with realistic figures for your AI plan.
- 5Run the review. Use /c-level-agents:caio-review with a clear description of your plan.
- 6Run the scripts if needed. Run the model, risk and cost scripts in the order listed in the workflow to get calculations for your data.
- 7Act on the verdict. Fix the gaps for SHARPEN, hold for BLOCK, and proceed on SHIP after checking the next steps.
What the caio-review skill makes the agent do
An agent skill is a folder containing a SKILL.md file that tells the agent how to handle a certain kind of task. This one, caio-review, is triggered by the command /c-level-agents:caio-review followed by a plan. The skill covers plans involving model selection, risk classification, cost economics, or AI hiring.
The agent works through six questions. They cover whether an evaluation set exists before shipping, the error-rate target and fallback, the EU AI Act risk tier, whether API, fine-tuning or building is the right path, the 12-month cost trajectory, and which role should be hired next. The output is a structured report ending in a SHIP, SHARPEN or BLOCK verdict, with next steps.
- Question 1: what the AI must do well and how you will measure it, including an eval set of 50-100 inputs minimum.
- Question 2: the hallucination or error-rate SLO and what happens when it is breached.
- Question 3: the EU AI Act tier (PROHIBITED, HIGH, LIMITED or MINIMAL) and whether a conformity assessment is required.
- Question 4: API, fine-tune or build, with an economic breakeven check.
- Question 5: the 12-month cost trajectory at expected scale.
- Question 6: which AI role to hire and whether prerequisite hires are in place.
Install the caio-review skill
Download the ZIP for the skill from this page and unzip it. You can install it in one of two places. For a personal setup, put the unzipped folder inside ~/.claude/skills/. For a single project, put it inside that project's .claude/skills/ folder.
If you use the Claude apps rather than a folder on disk, you can upload the skill instead. In the apps, go to Settings, then Capabilities, then Skills, and upload it there.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload
Check the requirements before you run it
The skill is written for Claude Code or a compatible agent that has the c-level-agents plugin installed. The SKILL.md refers to plugin paths under ${CLAUDE_PLUGIN_ROOT}, so the plugin must be present for those references to resolve.
The skill calls three Python scripts for its calculations: ai_risk_classifier.py, model_buildvsbuy_calculator.py and ai_cost_economics.py. You need Python to run them. These script files are not included in the material reviewed for this page, so confirm they exist in your plugin install before relying on the calculations. You also need JSON input files describing your use case and workload, such as use_case.json and workload.json, which the workflow uses as inputs.
- Claude Code or a compatible agent with the c-level-agents plugin
- Python, to run the referenced scripts
- JSON files describing the use case and expected workload
Trigger the skill with a natural request
Use the slash command with your plan in plain language. The skill is also meant to run when you describe one of the situations it covers, such as shipping an AI feature without an eval set, choosing between API, fine-tune and self-hosted, or classifying a use case under the EU AI Act.
Give the agent enough detail to work with. Name the feature, the users affected, whether any EU residents are involved, the expected request volume, and the model or vendor you are considering.
- Shipping a feature: "Review our plan to ship an AI support chatbot next month. We have no test set yet."
- Vendor decision: "Should we sign a three-year API contract or self-host a 70B model for our document tool?"
- Regulation: "Does our resume-screening feature fall under the EU AI Act?"
- Hiring: "Before we hire an ML engineer, check whether we need an AI engineer first."
/c-level-agents:caio-review <describe your plan>Read the report and act on the verdict
The report starts with the decision being made, then covers eval discipline, model selection, risk classification, cost economics, and org needs where they apply. It ends with a verdict: 🟢 SHIP, 🟡 SHARPEN or 🔴 BLOCK, followed by three concrete next steps.
Treat SHARPEN as a request to fix named gaps and rerun the review. Treat BLOCK as a hold on launch or commitment until the cited issue is resolved. The skill's routing notes also point to related reviews: cdo-review for training-data issues, gc-review for vendor contracts and liability, ciso-review for prompt injection and poisoning threats, and cfo-review for multi-year commitment TCO.
- SHIP: no blocking gaps found in the review.
- SHARPEN: fix the named gaps, then run the review again.
- BLOCK: do not launch or sign until the blocking issue is resolved.
Safety notes: scripts and network access
The skill runs scripts, so read them before you run them and only run scripts you trust. The skill's stated network use is none. The SKILL.md itself gives guidance and a template; the calculations come from the referenced Python scripts.
Treat the numbers with care. The skill's breakeven and TCO figures are rules of thumb unless the referenced scripts are run on real data. Its EU AI Act outline is rough, and the cost and timeline figures it quotes are generic. Have a lawyer confirm any real risk classification.
- Runs scripts: yes. Review the Python files before running them.
- Uses network: no, according to the reviewed material.
- Not legal advice on EU AI Act compliance.
- Cost and breakeven numbers are estimates unless you run the scripts on your own data.
Tips for better reviews
The quality of the review depends on the quality of the input. Write down your success criteria before you ask for a review, because the first question will ask for them anyway. Keep the JSON inputs realistic: real request volume, real token counts, and the true set of affected regions.
Run the three scripts in the order the workflow lists them, then compare the outputs with the report. If the report and the script output disagree, check the inputs first.
- Bring an existing eval set or a plan for building one.
- Use real traffic estimates in workload.json, not guesses.
- Include every region where EU residents are affected.
- Use the verdict as a checkpoint, then log it with /c-level-agents:decide if you use that command.
What people use it for
- Deciding whether an AI support feature has enough evaluation coverage to launch.
- Comparing a multi-year API contract with self-hosting before signing.
- Checking whether a hiring plan should start with an AI engineer or an ML engineer.
- Screening a new AI feature for a regulated domain such as employment or credit to see whether it needs a deeper legal review.
Frequently asked questions
Is the caio-review skill free?
The reviewed material identifies the skill as MIT licensed. It requires Claude Code or a compatible agent, the c-level-agents plugin, and Python to run the referenced scripts, so check those requirements for your own setup.
How do I install caio-review?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings, then Capabilities, then Skills.
What do I need to run caio-review?
You need Claude Code or a compatible agent with the c-level-agents plugin, Python to run ai_risk_classifier.py, model_buildvsbuy_calculator.py and ai_cost_economics.py, and JSON files describing your use case and workload. The script files are not included in the reviewed material.
Can caio-review give me legal advice on the EU AI Act?
No. The skill gives a rough tier outline for EU AI Act risk, and its cost and timeline figures are generic. A lawyer should confirm any real classification before you rely on it.
Are the cost and breakeven numbers accurate?
They are rules of thumb unless you run the referenced scripts on real data. Use realistic volume and token figures in your JSON inputs to get numbers worth discussing.
Does caio-review send my data over the network?
The reviewed material states that the skill does not use the network. It does run Python scripts, so review those scripts before running them.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use caio-review if you are a founder, product lead or team lead who needs a structured challenge before shipping an AI feature, signing a long vendor deal, or hiring for AI. Pick something else if you need binding legal advice on EU AI Act compliance or precise financial forecasts, since the skill's outputs are rough and need expert confirmation.
Download the skill