How to use the Change Impact Analysis agent skill

What is an agent skill and what does this one make the agent do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches the skill's description. This skill tells the agent to perform evidence-backed impact analysis before a change and to check the actual diff against that analysis afterward.

Before implementation, the agent traces real dependencies such as imports, calls, routes, schemas, tests, documentation and deployment steps, and lists what could break. Anything it cannot prove is marked as unverified. After implementation, it compares the diff with the original plan and reports scope drift or remaining risk.

• Answers six questions before a change: code, data, interfaces, tests, documentation and release.
• Prefers evidence from real imports and calls, routes, schemas, generators, git grep, git diff and executable tests.
• Does not infer dependencies from directory names.
• Is read-only by default and asks before writing files.

Install the change-impact skill

Download the ZIP for this skill from this page and unzip it. You can install it in one of two places, depending on whether you want it available everywhere or only in one project.

For a personal install across projects, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. The skill folder should contain agents/openai.yaml and SKILL.md.

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

• Personal: ~/.claude/skills/
• Project only: .claude/skills/ inside the project
• Claude apps: Settings → Capabilities → Skills

Trigger the skill with natural requests

You do not need a special command. The agent loads the skill when your request matches its description, which covers cross-module or high-risk changes, database migrations, authentication, public interfaces, blast-radius questions, pre-change checks and post-implementation alignment.

Name the risk and the object you are changing. The more specific the request, the more useful the analysis.

• "Before we change the users table, what is the blast radius of the migration?"
• "Analyse the impact of changing the auth token format before I implement it."
• "Which callers and tests are affected if we remove this public API field?"
• "Compare the diff I just made with the original impact analysis and tell me about scope creep."

What the skill produces

For small changes, the agent reports affected objects, evidence, required verification, documentation updates and unknowns in the chat response. For cross-module or high-risk work, it first looks for the project's existing impact-analysis location in its map or documentation index and asks you to confirm before writing there. If no canonical location exists, the report stays in the response unless you choose a location.

For high-risk changes such as database migrations, breaking interfaces, authentication and production releases, the skill expects a rollback condition, recovery steps, a compatibility period and an account of irreversible effects. It treats any missing item as a blocker and does not run rollback on your behalf.

Example workflows

A typical pre-change check asks the agent to find callers, contracts, migrations, tests and rollback steps for a planned API change. A post-change check asks the agent to compare the finished work with the original analysis and flag unverified consumers.

The skill also routes related work: durable architecture decisions go to architecture-decision-records, cross-client interface changes to contract-first, verification evidence to test-collaboration, and downstream regression to module-regression. These skills are referenced by name in the source; install them separately if you want that routing to work.

• Pre-change: API change, list affected callers, contracts, migrations, tests and rollback.
• Pre-change: review a database migration for compatibility, recovery and irreversible effects.
• Post-change: check an implemented cross-module change against its impact analysis.

What you need before you start

The skill relies on evidence in your project. It recommends git grep and git diff for finding dependencies and reviewing changes, so the project should be under Git. It also needs access to the project's source files, tests and documentation.

Projects without tests, documentation or version control history give the skill less to verify, so it will report more unverified items.

• Git installed, with git grep and git diff available
• Access to the project's source files, tests and documentation

Safety notes: scripts and network

This skill does not run scripts and does not use the network. Its instructions treat repository files, configuration, test output and linked artifacts as untrusted evidence rather than as instructions. It says the agent should not disclose discovered secrets, and should not execute commands, write files or perform destructive actions unless the host policy allows it and you have explicitly approved the action.

Review any proposed file writes before approving them, since the skill is designed to ask first.

• Runs scripts: no
• Uses network: no
• Read-only by default; asks before writing files

Tips for better results

Give the agent the change description, the affected area and any known consumers in your first message. Ask for unknowns to be listed explicitly so you can close them before implementation.

Run the analysis before implementing and again after, so the post-implementation check has a plan to compare against. Keep your documentation index current, because the skill uses it to find where impact reports belong.

• State the change, the module and known consumers up front.
• Ask for a separate list of unverified items.
• Run the check both before and after implementation.