Distilly is an MIT-licensed Python project that distills a person's messages, documents, interviews, and public sources into a reusable, source-grounded Person Profile — packaged as an Agent Skill — so AI agents and compatible bots can work with that person's experience, judgment, voice, and ways of working.

What it is

Distilly is the person-modeling layer for agents. It follows one workflow: messages, documents, interviews, and public sources go in; the distilly creator produces a Person Profile; that profile goes out to an Agent or a compatible Bot. It is built from observable experience, decision patterns, expression, and ways of working, and it states plainly that it does not claim to clone the person behind those materials. The project was formerly named Colleague Skill, or colleague-skill; the former name remains in use for search continuity and project history. The current release packages each profile as an Agent Skill, and the canonical creator Skill is named distilly, installed in a distilly directory. It lives in the AgentSkills ecosystem, carries the AgentSkills Standard badge, and requires Python 3.9+.

The concrete problem is context loss. A colleague quits, a mentor graduates, a teammate transfers, and the whole playbook and context leave with them. Instead of re-explaining a working style from scratch or hand-writing role prompts, a user supplies source material plus a description and receives a portable Person Profile that a supported host can install and invoke. Distilly is not limited to the colleague case: the same workflow serves colleagues, partners, family, old friends, idols, public figures, fictional characters, and the user themselves, across three person families.

Key capabilities

• End-to-end pipeline stated as Messages · documents · interviews · public sources → Distilly → Person Profile → Agent / Bot.
• One workflow builds source-grounded Person Profiles for three person families, then packages each profile as an Agent Skill.
• Canonical creator Skill named distilly, installed in a distilly directory.
• Native local Skill discovery documented for Claude Code, Hermes, OpenClaw, Codex, DeepSeek Harness, Pi, Grok Build, and OpenCode; Grok Bot is listed separately as a saved-Skill workflow preview.
• Community gallery where any skill or meta-skill links directly back to the contributor's own GitHub repo, with no middleman.
• Documentation translations in Chinese, Spanish, German, Japanese, Russian, Portuguese, and Korean.
• Discord community plus a published technical report on arXiv (2605.31264) covering the earlier COLLEAGUE.SKILL work.

Who uses it and how

• Teams who lose a colleague to resignation or transfer, and want to keep the playbook and context as an installable Skill rather than reconstructing it later.
• Individuals distilling a partner, family member, old friend, or themselves, holding on to how it felt to work or talk with that person.
• People working with authors, idols, or thinkers they will never meet, asking what those figures would say about a given question, using public sources as input.
• Community contributors building and publishing Skills: the gallery has reached 215 skills contributed by 165 people, with more than 100,000 stars across the skill cards.
• Host users on Claude Code, Codex, DeepSeek Harness, and other supported environments who install profiles through native local Skill discovery.

Getting started

Install the canonical creator Skill named distilly in a distilly directory, following the README's Install section. Python 3.9+ is required, and supported hosts discover the Skill locally once it is in place.

How it compares

No list of paid products this project replaces is provided, and none of the comparable tools named in the facts are direct substitutes. Distilly positions itself as the person-modeling layer that sits across the hosts it names — Claude Code, Codex, DeepSeek Harness, Hermes, OpenClaw, Pi, Grok Build, and OpenCode — rather than as a competing host, so it stands alone in this registry.

When to use it — and when not to

A self-hoster runs the Python 3.9+ toolchain and maintains a Skill directory inside a supported host; the facts name no hosted service, database, or storage component to operate. Anyone expecting a verbatim replica of a person should not pick it, because the project explicitly limits its output to a source-grounded profile and disclaims cloning the person behind the materials. The project is also in active flux: it was renamed from Colleague Skill, the creator Skill naming changed as recently as 2026-08-24, the roadmap is open, and 37 issues remain unresolved.