How to use the Profile Manager agent skill (Profile)

What is an agent skill, and what does Profile do?

An agent skill is a folder of instructions that an AI agent such as Claude Code can load to handle a specific kind of task. The agent reads the skill's instructions when a request matches it.

Profile works on one file: the LifeOS User Model, PRINCIPAL_IDENTITY.md. It has three operations. View shows the current profile in full. Edit changes only the sections you ask about. Clear resets the file to a blank template after making a backup and getting your explicit confirmation.

• View: shows every section of the file, in file order, with its review dates.
• Edit: updates identity, goals, constraints or preferences you request.
• Clear: resets the contents to a blank template; the file itself is kept.

Install the Profile skill

Download the ZIP from this page and unzip it. Put the Profile folder in your personal skills directory at ~/.claude/skills/, or in a project's .claude/skills/ folder if you want it limited to that project.

If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.

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

Check the requirements before you start

The skill assumes a specific LifeOS directory layout. The identity file must exist at ~/.claude/LIFEOS/USER/PRINCIPAL/PRINCIPAL_IDENTITY.md, and the agent must load it through the CLAUDE.md @-import at startup. The skill was written for Claude Code or a compatible agent with that loading behaviour.

Git is optional. The user directory may be a private git repo, which helps with restores. The source also mentions Node/TypeScript hooks and a HeaderMirror.ts tool for header sync, but those files were not included in the files reviewed, so do not assume they are present.

Trigger the skill with natural requests

You do not need a special command. Ask in plain language and the agent routes the request. Trigger phrases in the skill include profile, show profile, view profile, who am I, my profile, edit profile, update profile, change profile, clear profile, delete profile, reset profile, and user model.

Requests about settings.json or CLAUDE.md are not for this skill. The skill says to use the update-config skill for those.

• "show my profile" runs View.
• "update my profile — add that I prefer metric units" runs Edit on the Preferences section.
• "clear my profile" runs Clear, with a backup and confirmation first.

Understand what each operation changes

Edit changes only the sections you ask about and preserves other data. It sets last_updated and last_updated_by. It never sets last_reviewed, so the freshness grade stays honest. Specific requests are applied directly; an open-ended request like "edit my profile" gets the current contents and a question about what to change.

Clear backs the file up to ~/Downloads/PRINCIPAL_IDENTITY.<timestamp>.md, checks the backup is non-empty, then overwrites the file with the blank template after you confirm. It removes last_reviewed so the file shows as unreviewed. Changes take effect in the next session, because the file loads at startup.

Safety notes: scripts, network and sensitive data

The skill does not run scripts and does not use the network, according to the review. Its risks are about the data it handles.

The file is loaded into every session, so anything you put in it is in context every time. The skill only records health, relationship or financial details when you explicitly ask. Keep sensitive material out of the profile unless you want it loaded every session. The skill also keeps executable directives out of the file; those belong in CLAUDE.md or the system prompt.

• Clear does not delete the file, but it does overwrite its contents; check the backup in ~/Downloads first.
• Facts kept only in frontmatter may not reach the model unless the header-mirror block carries them.

Tips for a profile that stays useful

Keep entries short and specific, since the whole file is loaded every session. After a change, start a new session or reload context so the agent sees it. If you want to check what the agent knows, ask to view the profile rather than trusting memory.

• Make one clear change per request, so the edit is easy to review.
• Ask for View before Edit when you are unsure what the file already says.
• Keep a git history of the user directory if you use one, so you can restore old versions.