How to use the REA CI and Tests Audit agent skill

What is an agent skill?

An agent skill is a folder of instructions and supporting files that an AI agent loads when a task matches the skill's description. Instead of re-explaining your project's process every time, you give the agent a skill that describes how to work on a specific kind of task.

This skill is named rea-ci-tests-audit and is published under the MIT licence. Its description says it is for suite or workflow investigations and efficiency changes, not ordinary feature testing or diagnosing a single test failure.

What the skill makes the agent do

When the skill is active, the agent is directed to establish a baseline first: it pins the commit it inspects, records the exact commands, toolchain and prerequisites it uses, and keeps raw audit notes outside the tracked tree, normally in a task-specific /tmp directory.

It then traces how CI actually runs: workflow triggers, changed-path selection, job dependencies, package scripts and test discovery. To claim a problem, the agent must reproduce it against real behaviour. For a suspicious green test, it perturbs the behaviour in an isolated copy and checks that the test fails. For speed claims, it measures equivalent completed work before reporting a speedup.

• Traces workflow triggers, test discovery and report gates to the commands that really run
• Reproduces suspected problems and checks fixtures against current requirements and actual entrypoints
• Deliberately breaks code to confirm a test would notice
• Measures timing and memory before claiming a speedup
• Does not edit code, push or merge unless you explicitly authorise it

Install the skill from the ZIP

Download the ZIP file for the skill from this page and unzip it. You have two options depending on how you use your agent.

For a personal setup, unzip the skill into ~/.claude/skills/ so it is available across your projects. For a single project, unzip it into the project's .claude/skills/ folder so it applies only to that repository.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/
• Project only: .claude/skills/ inside the repository
• Claude apps: Settings → Capabilities → Skills, then upload

Check the requirements before you start

The skill only works properly inside a checkout of the REA repository. The source expects the checkout to contain AGENTS.md, docs/testing.md and CONTRIBUTING.md, because the skill tells the agent to follow those files for current commands and the pinned toolchain.

The skill does not list the toolchain or test prerequisites itself; it defers to the repository's own documentation. Some optional providers such as Ghidra, LLDB or a browser are used by certain tests, which skip when they are absent. The skill runs scripts and uses the network, so review what it will run before you start.

• A checkout of the REA repository with AGENTS.md, docs/testing.md and CONTRIBUTING.md
• The pinned toolchain and test prerequisites named in the repository's documentation
• Optional: Ghidra, LLDB or a browser, for tests that skip without them
• The skill runs scripts and uses the network

Trigger the skill with natural requests

You do not need special syntax. Describe the investigation you want in plain language and mention the CI or test context. The skill's description covers suite and workflow investigations, flaky or slow runs, and efficiency changes.

Keep the scope explicit. If you only want findings, say so; the skill says investigation alone does not authorise edits.

• Check whether our CI would catch a broken fixture in the REA test suite, but do not change anything yet.
• This CI run looks suspiciously green. Trace the workflow triggers and test discovery and tell me if a required test was skipped.
• Find out which slow test families in REA CI could be made faster without removing any assertions. Measure before claiming a speedup.
• Audit the fixtures for the admission checks against current requirements and report any drift.

Review the results and keep control

The skill asks the agent to report each confirmed issue with the exact revision, reproduction, consequence, proposed correction and validation limits. Performance is to be reported with workload and measurements rather than a single percentage.

The skill is designed for a human to review scope and any proposed merge. Do not use it for unattended CI changes, because it can lead to edits, reruns and full local suite runs.