Search skills/

Distilly Claude skill

Skill that distills a person's messages, documents and public material into a reusable Person Profile packaged as an agent skill.

titanwings/distilly 1 skill Memory & contextSkill & agent tooling
Our verdict Recommended with caveats

A substantial toolkit with tests, but it processes other people's private messages, so consent and data handling matter more than the code.

4/5 Depth expertise shipped ! Security Broad access 16d Maintenance Updated 16 days ago 26K Adoption stars OK License MIT, commercial OK

What Distilly does

Distilly, formerly Colleague Skill, builds a Person Profile from material you provide: chat history, documents, interviews or public sources. The profile captures observable experience, decision patterns, expression and ways of working, and is packaged as an Agent Skill so a supported host can install and invoke it. The README frames uses from a departed colleague's playbook to a mentor, a favorite author, a fictional character or yourself, and says it does not claim to clone the person. The canonical skill is a short SKILL.md (about 4 KB) that also handles first-run setup of an owner profile so a personal agent knows who it works for. Behind it is a larger toolkit: roughly 40 scripts, 13 reference files and 15 test files, with CI. Data collection covers Lark through its API, DingTalk through the browser, Slack through an admin-installed bot, and manual uploads of exports. Credential setup and per-host install paths are in a separate install guide. The project documents translations in seven languages and had passed 20,000 stars by August 2026.

Good for

  • Capturing a colleague's working knowledge before they leave, with their agreement
  • Building a persona skill from public writing by an author or thinker
  • Teams already on Lark, Slack or DingTalk

Not for

  • Profiling people from private messages without their consent
  • Users who want a simple prompt-only skill

What's inside depth 4/5

A short 4 KB SKILL.md backed by about 40 scripts, 13 reference files, 15 test files and CI in a 133-file repo.

Skills
1
Reference docs
13
Scripts
42
Tests / evals
15
Templates
0
Avg SKILL.md
3.8 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • Lark app bot, Slack bot with admin install, or exported chat files as the data source
  • Python for the collection and research tools
  • Optional audio transcription for interview material

Security review

Broad access

This skill asks for more access than most, or sends data to another service. That is usually a design choice rather than bad intent, but you should know before you install it.

  • Pre-approves broad shell commandsFrontmatter allowed-tools: Read, Write, Edit, Bash pre-approves unrestricted Bash (and file writes) for the whole skill.SKILL.md:7

Distilly builds 'person skills' from material about a colleague, mentor or public figure, including chat history pulled from Feishu, DingTalk and Slack with API tokens you provide, and and, if you choose that option, from Feishu pages via your own logged-in Chrome profile. The skill pre-approves unrestricted Bash, writes generated skills and optional profile blocks into your agent's user folders, and runs npx feishu-mcp unpinned. The data it gathers is other people's private messages, so check consent and where outputs are saved.

Worth knowing

  • browser_session_optionalOne of two documented Lark options (the user picks 'Browser' or 'MCP'; the browser option is for internal docs without App permission): tools/feishu_browser.py launches Playwright Chromium on your real Chrome profile directory (Default profile) to reuse your logged-in session, with automation-detection flags turned off, and reads docs, wiki and chat messages with it.tools/feishu_browser.py:47
  • Makes network requestsCollectors call the Feishu, DingTalk and Slack APIs to download chat history, docs and mail of a named colleague, and research tools call xquik.com for public X posts, yt-dlp for subtitles and local/remote speech-to-text.tools/feishu_auto_collector.py:69
  • Needs an API key for its own serviceFeishu app secret/user token, DingTalk app key/secret, Slack bot token and XQUIK_API_KEY are read from env, the OS keychain or a 0600 ~/.distilly config and sent only to their own platform APIs; the docs say to enter them in your own terminal, not in chat.tools/credentials.py:25
  • Runs remote code at an unpinned versionThe Feishu MCP client runs npx -y feishu-mcp --stdio at whatever version npm serves, passing your Feishu credentials in its environment.tools/feishu_mcp_client.py:130
  • Changes files outside your projectInstallers write generated person skills into the host's skills folder (e.g. ~/.claude/skills), and owner_profile.py can insert a block into ~/.claude/CLAUDE.md or other agents' user files; owner_sources.py lists names of your MCP servers and skills from ~/.claude.json and ~/.codex/config.toml (names only).tools/owner_profile.py:62
  • Installs packages or toolsOptional features need pip packages (requests, playwright with Chromium, keyring, faster-whisper) and yt-dlp; the npm package is published to GitHub Packages via @titanwings registry.INSTALL_EN.md:109
  • Relies on a CLI or package we could not reviewRelies on yt-dlp, npx feishu-mcp, Playwright and whisper backends that are not in the repo.tools/research/download_subtitles.sh:9

What we checked

The repository (133 files) on Oct 29, 2026. Not a full review: we read 4 of 46 code files; it relies on code outside the repository: npm package feishu-mcp, Playwright/Chromium, yt-dlp, faster-whisper/openai-whisper, keyring and requests; Most of the 46 listed code files (the Feishu/DingTalk/Slack collectors, email_parser, skill_writer and the tests) were pattern-searched and excerpted, not read in full; credentials.py, feishu_browser.py, owner_sources.py, owner_profile.py and feishu_mcp_client.py were read in part.

NVIDIA SkillSpector static scan: 119 pattern hits (Privilege Escalation 43, Data Exfiltration 17, Dangerous Code Execution 14, Supply Chain 12, Data Flow 10, Rogue Agent 5). Most hits are the credential/keyring handling in tools/credentials.py and its tests, env copies in test helpers, 'rm' lines in prompt flows and test scripts, the GitHub Packages registry line in .npmrc, and scanner notes about partially read prompt files. The sensitive items are real by design: browser-profile reuse, chat collectors and shell pre-approval.

A review, not a certification. Skills change after we check them — read the files before installing.

License

MIT — commercial use allowed. Check workplace policies and privacy law before processing colleagues' messages.

Can it run unattended?

Not really. Profile building depends on you supplying material and reviewing the output.

Alternatives to Distilly

memU

NevaMind-AI/memU

Memory layer that mines agent session logs into a shared markdown wiki and re-injects it across Codex, Claude Code, Cursor and others, locally or via memU Cloud.

With caveats Broad access Depth 4/5

task-observer (One Skill to Rule Them All)

rebelytics/one-skill-to-rule-them-all

A meta-skill that watches your sessions, logs corrections and repeated work, and proposes improvements to your other skills for you to approve.

Recommended Check first Depth 4/5

memsearch

zilliztech/memsearch

Cross-agent persistent memory for Claude Code, Codex, OpenClaw, OpenCode and DeepSeek Harness: Markdown files as the source of truth, with a Milvus search index.

With caveats Broad access Depth 3/5

FAQ

What is Distilly?

Distilly turns messages, documents and public sources about a person into a source-grounded Person Profile that an agent or bot can use. It was previously called Colleague Skill.

How do I install Distilly?

The canonical skill must be installed in a directory named distilly. The README points to an Install Guide (INSTALL_EN.md) for host paths, Windows notes, generated-profile installation and credential setup.

Is it ok to use Distilly on colleagues' chats?

That is a legal and ethical question the tool does not answer for you. Get the person's agreement, follow your employer's data rules, and avoid sensitive private conversations.

Distilly vs a memory skill like memU?

Distilly builds a profile of another person from source material. A memory tool such as memU records your own agent sessions and reuses them across agents.

Sources: GitHub repository (stars, license, file tree, last push Oct 13, 2026). Review based on the README and SKILL.md files as published.