Distilly Distills a Person's Playbook Into a Reusable Agent Skill

Your best colleague leaves and takes the playbook with them. Distilly — 25k stars, formerly Colleague Skill — builds a Person Profile from the material a person left behind and packages it as an Agent Skill your agent can install and invoke. The README draws its own boundary up front: the profile is source-grounded and does not claim to clone the person behind the patterns.

Why This Skill Matters

One workflow covers three person families, each with its own source strategy and profile structure:

• colleague — coworkers, mentors, upstream and downstream partners: a Work Skill plus Persona built from material-derived technical standards, workflows, expression, and workplace behavior.
• relationship — partners, family, old friends: expression patterns, emotional triggers, conflict patterns, and repair patterns.
• celebrity — public figures and fictional characters: a six-dimension research toolchain covering works, interviews, decisions, expression, external evaluations, and timeline.

Every generated profile shares a Persona base, with family-specific modules layered on top; for the colleague family, that base is a six-layer stack — hard rules, identity, expression, decisions, interpersonal style, and corrections. Profiles also evolve: append new files and the analysis merges into the relevant sections without overwriting existing conclusions; say "they wouldn't do that, they'd be more direct" and the note lands in the Correction layer immediately; every update auto-archives so you can roll back.

Installation

The README's agent-facing install is a single instruction:

Install Distilly from https://github.com/titanwings/distilly, then verify that this host can discover it.

For humans, clone it into the skills directory your host scans:

git clone https://github.com/titanwings/distilly <DISTILLY_SKILL_DIR>

Claude Code, Hermes, OpenClaw, Codex, DeepSeek Harness, Pi, Grok Build, and OpenCode have native local Skill discovery documented. Grok Bot is listed separately as a saved-skill workflow preview — direct repo installation is not yet verified there. The install guide covers host paths, migration, Windows, and credential setup.

Real Workflow: Keep a Departing Teammate's Judgment

1. Collect the material the README ranks highest for colleagues: their own long-form writing (design docs, review comments) and decision-making replies, ahead of casual group chat.
2. In your agent, say:

Use Distilly to create a Person Profile for <person>.

3. Choose the colleague family, attach the material and a description — every field is optional — then review the result and generate.

The result is an Agent Skill named {character}-{slug}. Ask it to review an API design and it answers from the distilled standards, workflows, and preferences rather than generic best practice.

Real Workflow: Celebrity Research

Choose the celebrity family and the six-dimension pipeline organizes works, interviews, decisions, expression, external evaluations, and timeline. Source priority here: first-person books, blogs, and long interviews first, then decision records such as launches, commits, and Q&A, then verified first-person short-form posts, with third-party commentary last.

Tips

• Data sources are flexible: Lark auto-collection via API, Slack through an admin-installed bot (the free plan is limited to 90 days), WeChat history exported with WeChatMsg or PyWxDump, PDFs and screenshots, .eml or .mbox email, and pasted markdown.
• Feeding more material later is the intended flow — append and let the delta analysis merge.
• Public X research runs through a metered third-party service called Xquik.

When Not to Use This

The README says it outright: this is still a demo version. Quality is bounded by source quality — thin or second-hand material produces a thin profile — and the output models observable patterns, not the person. Skip it wherever a wrong voice would cost more than a useful one.

See the leaderboard for more skills.