About-Me Skill Creator

Create a private, user-specific about-me skill. The resulting skill helps every agent in a session calibrate to the person running it; this creator is not itself their profile.

Safety Boundaries

• Treat every answer as personal context. Never add credentials, tokens, private keys, or information the user would not want read aloud.
• Do not copy repository-specific instructions into the generated skill. Those belong in repository files such as AGENTS.md.
• Include durable facts only. Omit task-specific details and dated priorities unless the user explicitly wants a maintained Current Focus section.
• Keep the profile concise: normally 40 to 80 lines. Delete empty or generic sections.

Wizard

Run this as a conversation, not a questionnaire. Ask one small group of questions at a time, show the draft as it develops, and let the user skip any section.

1. Select a Mode

Ask whether the user wants to:

1. Create a new profile.
2. Review or refresh an existing profile.
3. Draft a profile from material they provide, such as a CV, bio, README, or existing instructions.

For a refresh, read the existing profile first and preserve valid, still-durable facts. Remove anything stale rather than accumulating contradictory entries.

2. Select a Destination

Ask where to create the generated skill. Offer these choices:

• ~/.claude/skills/about-me (default)
• ~/.agents/skills/about-me
• A custom absolute path

Resolve ~ before writing. State the final <destination>/SKILL.md path and ask for confirmation before replacing any existing file. Create missing parent directories when the current environment permits it. If the agent cannot write there, provide the complete generated file and the exact destination instead.

3. Gather the Profile

Work through these topics in order. Ask for the reason behind a preference whenever it would help an agent generalize beyond the immediate example. Challenge vague input: "be professional" and "write good code" do not calibrate behavior.

1. Working style: desired directness, answer length, reasoning depth, disagreement, uncertainty, and feedback that has repeatedly missed the mark.
2. Autonomy: what an agent can decide, what requires prior approval, and what is forbidden without explicit instruction.
3. Background: role, fluencies to assume, and areas where extra explanation helps.
4. Tooling and constraints: editors and agent clients, stack, platform, budget, offline, compliance, approval, or licensing limits.
5. Hard rules: a short list of genuinely non-negotiable practices.
6. Current focus: optional, dated, and included only when the user intends to maintain it.

Ask what is already documented before re-asking questions answered in user-provided material. A correction the user has had to give twice is especially valuable: turn it into a specific, durable instruction.

4. Draft and Review

Draft a standalone SKILL.md using the structure below. Replace all bracketed text with the user's facts and omit sections that do not add signal.

---
name: about-me
description: Durable personal context about the person running this session — their background, tooling, constraints, and preferences for how agents should work with them. Load at the start of every session and whenever calibrating tone, autonomy, technical depth, or workflow.
license: MIT
---

# About Me

## Working Style

- [Specific preference, ideally with its reason.]

## Autonomy

- **Decide without asking:** [Boundary.]
- **Ask first:** [Boundary.]
- **Never without explicit instruction:** [Boundary.]

## Background

- [Role, fluencies, and areas needing explanation.]

## Tooling and Environment

- [Tools, stack, platform, and constraints that change decisions.]

## Current Focus

- [Optional dated focus item.]

## Hard Rules

- [Short non-negotiable rule.]

Before writing, ask: "If an agent knew only this file, would it behave noticeably differently?" Replace generic statements with concrete corrections or remove them. Confirm the destination and final content, then write <destination>/SKILL.md.

Maintenance

Tell the user that their generated skill is private and should not be committed to a shared repository. When they give the same correction twice, refresh this profile and add the missing durable instruction.